DEV Community: Saptarshi Paul

How ChatGPT and Gemini Choose Which Brands to Recommend: A Pipeline Walkthrough

Saptarshi Paul — Fri, 02 Oct 2026 05:50:02 +0000

Ask ChatGPT which payment gateway to use and you get a confident, well reasoned answer. It sounds like the model knows the market. It doesn't, at least not from training. The answer comes from a pipeline: the model searches the web, picks top results, reads passages, reasons over them, then writes. The language model is one stage in that chain, and not the one that decides whether your brand is in the running.

The Short Version

The language model is a frozen snapshot. Its knowledge stops months before you ask.
The apps hand the model tools. For product questions it makes at least two tool calls: a web search, then a fetch of pages worth reading.
Your prompt gets rewritten into several search queries. Ordinary search ranking decides which pages reach the model at all.
The model reads passages, not whole pages, then weighs them against what it already believed.
Every step is a filter. To come out the other end, a brand has to be indexed, fetchable, quotable and backed up by other people.

The model on its own cannot know about you

A large language model is trained in two phases. Pretraining feeds it a huge slice of the public web so it learns patterns. Fine-tuning then teaches it to be helpful and follow instructions. Both phases finish on a fixed date, and after that date the weights do not change. Nothing new gets in.

That date is the knowledge cutoff, and it usually sits further back than people assume. GPT-5 shipped in August 2025 with a September 2024 cutoff. Gemini 2.5 Pro has a January 2025 cutoff. Newer models narrow the gap, but there is always a gap. A model released this month still knows nothing about the pricing page you shipped last week.

There is a second, quieter problem. Inside the cutoff, what the model holds about your brand is a statistical impression rather than a record. It has seen your name near some words more often than near others. It cannot reliably tell your 2024 feature set from your 2025 one. Ask a model with no tools for the best CRM for a 10 person agency, and you get a confident answer built from patterns that are a year old. It will not flag that for you.

The model's memory is out of date. GPT-5 was 11 months behind on launch day and 2 years behind today.

So the product teams did the obvious thing. They stopped asking the model to know things and started asking it to look things up.

Tools: the model asks, software answers

The mechanism is called tool calling, and it is simpler than the name suggests. The model cannot open a web page. All it can do is produce text. So the app gives it a short menu of actions written in plain language: search the web, open this URL, find this phrase on the page. When the model wants one of those, it writes a small structured request instead of an answer. The app catches that request, runs the real search or fetch, and pastes the result back into the conversation. Think of it as a colleague handing the model a printout.

The model reads the printout, thinks about what it says, and writes the answer. If a page looks worth opening, it asks for that too before writing.

This is the whole idea. A model that cannot see the internet becomes a model that sends errands and reads what comes back. OpenAI, Google, Perplexity and Anthropic all run some version of this. Google's developer docs for Gemini lay it out in five stages: the model looks at the prompt, decides whether a search would help, writes one or more queries, runs them, then works the results into a response. ChatGPT's browsing tool does the same job with different plumbing.

The rest of this post goes through those steps one at a time, because each step throws pages away. The pages that survive are the ones that get your brand named.

Step 1: the model decides whether to search

Not every prompt triggers a search. Explain compound interest usually does not. Best project management tool for a remote team of 12 almost always does, because the model has learned that product questions go stale. In ChatGPT a small decision layer makes this call before any search runs. In Gemini the model makes it as part of its reasoning.

For anyone trying to get a brand named, the practical point is short. Anything that names a category, a price, a comparison or the current year goes to the web. Your brand's fate on that prompt depends on what the web says today, not on what the model absorbed in training. Which is a relief, because you can change the web.

Step 2: your question becomes several queries

The model does not paste your prompt into a search box. It rewrites it. One question becomes a set of narrower searches, each aimed at one facet of what you asked. Google calls this query fan-out and describes it as issuing multiple related searches concurrently across subtopics and multiple data sources. ChatGPT does the same thing, it just has not named it.

Take which payment gateway should a UK subscription business use. A plausible fan-out looks like this:

payment gateways for subscription billing UK
Stripe vs GoCardless vs Adyen recurring payments
payment gateway fees comparison 2026
SCA and 3D Secure support subscription payments
best payment gateway reviews small business UK

Notice that your page does not need to rank for the original question. It needs to rank for at least one of the sub-queries, and the narrow ones are the easier target because fewer strong pages compete for them.

ChatGPT's shopping answers go one step further. Profound looked at about 200,000 shopping prompts and found a second fan-out layer that turns the prompt into product-specific sub-queries before any candidate is picked. Same pattern: many narrow searches rather than one broad one.

Step 3: a search engine picks the candidates

Each of those queries runs against a real search index. For ChatGPT that is a mix of providers with Bing as the main one, plus whatever OpenAI's own crawler, OAI-SearchBot, has indexed. For Gemini, AI Overviews and AI Mode it is Google's index, its Knowledge Graph and, for products, its Shopping Graph. Perplexity runs its own index and ranker.

What comes back is a ranked list: title, URL, snippet, date. Not the page, just the listing. The model sees maybe 10 results per query, so the search engine's ranking decides which pages even exist as far as the model is concerned.

This is where classic SEO still earns its keep. An early study of ChatGPT search found 87% of its citations matched Bing's top results. The overlap has loosened since, but the principle holds. If no search engine ranks your page for any of the fan-out queries, you were never a candidate.

Two things follow. Being invisible on Bing now hurts you in ChatGPT in a way it never did in Google. And a robots.txt rule that blocks OAI-SearchBot or Google's crawler removes you from the pool entirely, however good the page is.

Step 4: the model opens the pages and reads

Now the second tool call. The model looks at the listings and picks the ones worth opening. For those, the app fetches the live page. OpenAI does this with a separate fetcher, ChatGPT-User, which only runs when someone's question needs it. Google reads from its own cache and from the live web.

The model does not read the whole page. Fetched pages get stripped to plain text, and the model reads them in passages, looking for the bit that answers the sub-query it is working on, then moves on. A 3,000 word page contributes the two paragraphs that match. If they match.

That changes how I think about writing a page:

A claim buried in the fourth paragraph of a section is easy to miss. The same claim in the first sentence under a clear heading is easy to lift.
A sentence that only makes sense with the paragraph around it cannot be quoted on its own. Fees start at 1.4% plus 20p per transaction can. Our fees are competitive cannot.
Pages that only render their main content after JavaScript runs often come back nearly empty to the fetcher. Worth checking yours.

You can watch this step happen. In ChatGPT, expand the Searched the web line above an answer and you see the queries it ran and the pages it opened. Gemini shows the same in its sources panel. Doing this for your own category is the fastest education in GEO I know.

The four pages behind one answer: two of the vendor's own, two from its rivals.

Step 5: the model reasons over what it read

With a stack of passages in front of it, the model does what it is actually good at. It compares them. It notices when four independent pages make the same claim and when only one does. It weighs a vendor's own page against a review site against a Reddit thread. And it blends all that with what it already believed from training, which still counts. A brand the model has seen a thousand times starts with a prior in its favour. There is no way around that except to be seen more.

Newer reasoning models make this step longer and more deliberate. They plan, search, read, notice a gap, search again. GPT-5 Thinking, Gemini's Deep Research and Perplexity's Pro Search all run several rounds before writing. The number of tool calls behind one answer can run into the dozens.

What survives this stage is usually not the loudest page. It is the claim that is specific, checkable, and repeated by other sources. Vague positioning gets a brand listed. A number, a named feature and a third party saying the same thing get it listed first.

There is data for this. In Profound's shopping study, the product in the top slot had a median of 787 reviews against 352 for the rest. Nobody told the model to count reviews. It reads a review count as other people agreeing, and acts on it.

Step 6: the model writes, and cites

Finally it writes a conversational answer and attaches citations. Those citations are the pages that made it through all five steps: ranked by a search engine, chosen for opening, read, and found useful. They are the audit trail of the whole process, and I read them that way.

A citation is also the only link a reader might click, and very few do. Pew tracked 900 US adults and found people clicked a source inside an AI Overview on about 1% of visits. So being cited matters because it means you were in the answer, not because it sends you traffic. It mostly does not.

Why people trust these answers

Put the steps together and what comes out is a synthesis of 10 to 30 pages, cross-checked, with the disagreements smoothed over and a recommendation on top. That is a better first draft of a decision than a page of blue links, and people behave accordingly.

The same Pew study found that when an AI summary appears, people click a traditional result on 8% of searches, against 15% when there is no summary. ChatGPT passed 900 million weekly users in February 2026 and kept growing. Google's AI Mode, built on exactly these steps, is now how a large share of searchers meet a product category for the first time.

None of that means the answers are always right. The process inherits whatever the top search results say, and the top results for a commercial query are not exactly neutral. But the answers are consistent, confident, and they arrive in one place, and that turns out to be enough to move where the decision gets made. OpenAI's own retreat from in-chat checkout in March 2026 shows the shape of it: people decide in the chat and buy on the site.

Which pages survive the steps

Step back and the steps are a series of filters, each one removing pages. What the studies say about the survivors is roughly this.

Sources are concentrated, and they move. Semrush tracked 230,000 prompts over 13 weeks and watched Reddit's share of ChatGPT citations fall from about 60% to about 10% in a single quarter, while PR Newswire, Forbes and Medium rose. The set of trusted sources per category is small and it does not sit still. Whatever the model cited last month is not a safe bet for next month.

Each engine has its own habits. In the same period Google's AI Mode leaned on LinkedIn, YouTube and Reddit, and cited Wikipedia in around 2% of answers. ChatGPT cited Wikipedia far more. Being visible in one engine tells you little about the others, because the search index behind each one is different.

Fresh, specific pages beat broad, old ones. The fetcher reads passages, so a focused page that answers one sub-query head-on tends to beat a general page that mentions the topic in passing. Recency is right there in the listing the model sees, so it uses it.

Third parties carry more weight than you do. Your own page tells the model what you claim. A review site, a forum thread or a comparison article tells it whether anyone agrees. The model treats agreement as evidence, which, to be fair, is what most of us do too.

Putting the work in order

The steps themselves dictate the order:

Be in the index. Google's and Bing's. Let OAI-SearchBot and ChatGPT-User through.
Be readable. Server-rendered text, clear headings, one claim per sentence, numbers where numbers exist.
Be backed up. Get the same specific claims onto pages you do not own.
Be measured. Run the prompts your buyers ask, weekly, on every engine you care about, and log which pages got cited and where your brand landed in the list.

The fourth one is the step most teams skip, and it is the only one that tells you whether the first three worked. Doing it by hand for 20 prompts across four engines takes me about an afternoon a week. Lumirank exists to turn that into a dashboard, but I would still do the manual version once. Open ChatGPT, ask the question a buyer would ask, expand the search panel, and read the list of pages it chose. Everything in this post is visible in that list.

Reference

How ChatGPT, Gemini & Other AI Choose Brands to Recommend

How ChatGPT and Gemini Choose Which Brands to Recommend: A Pipeline Walkthrough

Saptarshi Paul — Thu, 01 Oct 2026 16:06:30 +0000

The Short Version

The language model is a frozen snapshot. Its knowledge stops months before you ask.
The apps hand the model tools. For product questions it makes at least two tool calls: a web search, then a fetch of pages worth reading.
Your prompt gets rewritten into several search queries. Ordinary search ranking decides which pages reach the model at all.
The model reads passages, not whole pages, then weighs them against what it already believed.
Every step is a filter. To come out the other end, a brand has to be indexed, fetchable, quotable and backed up by other people.

The model on its own cannot know about you

The model's memory is out of date. GPT-5 was 11 months behind on launch day and 2 years behind today.

So the product teams did the obvious thing. They stopped asking the model to know things and started asking it to look things up.

Tools: the model asks, software answers

The model reads the printout, thinks about what it says, and writes the answer. If a page looks worth opening, it asks for that too before writing.

The rest of this post goes through those steps one at a time, because each step throws pages away. The pages that survive are the ones that get your brand named.

Step 1: the model decides whether to search

Step 2: your question becomes several queries

Take which payment gateway should a UK subscription business use. A plausible fan-out looks like this:

payment gateways for subscription billing UK
Stripe vs GoCardless vs Adyen recurring payments
payment gateway fees comparison 2026
SCA and 3D Secure support subscription payments
best payment gateway reviews small business UK

Step 3: a search engine picks the candidates

Step 4: the model opens the pages and reads

That changes how I think about writing a page:

A claim buried in the fourth paragraph of a section is easy to miss. The same claim in the first sentence under a clear heading is easy to lift.
A sentence that only makes sense with the paragraph around it cannot be quoted on its own. Fees start at 1.4% plus 20p per transaction can. Our fees are competitive cannot.
Pages that only render their main content after JavaScript runs often come back nearly empty to the fetcher. Worth checking yours.

The four pages behind one answer: two of the vendor's own, two from its rivals.

Step 5: the model reasons over what it read

Step 6: the model writes, and cites

Why people trust these answers

Which pages survive the steps

Step back and the steps are a series of filters, each one removing pages. What the studies say about the survivors is roughly this.

Putting the work in order

The steps themselves dictate the order:

Be in the index. Google's and Bing's. Let OAI-SearchBot and ChatGPT-User through.
Be readable. Server-rendered text, clear headings, one claim per sentence, numbers where numbers exist.
Be backed up. Get the same specific claims onto pages you do not own.
Be measured. Run the prompts your buyers ask, weekly, on every engine you care about, and log which pages got cited and where your brand landed in the list.

Reference

How ChatGPT, Gemini & Other AI Choose Brands to Recommend

10 Promptwatch Alternatives Compared on a Real Workload: Pricing, Seats, and Limits

Saptarshi Paul — Thu, 01 Oct 2026 06:39:04 +0000

Say you run marketing at a UK payment gateway. There are three of you, and you want to know whether ChatGPT, Perplexity, Gemini, and Google AI Overviews name your company when a founder asks for the best payment gateway for a UK SaaS business. Promptwatch is a common place to start, and it is a good tool. Lining it up against alternatives is harder than it looks because every vendor lists the same features. What actually separates them is what one concrete job costs and how often each tool runs it. This post prices one common setup on every tool: 100 prompts on 4 AI apps, run every day. On Promptwatch that costs $245 a month, or $579 once a third person on your team needs a login.

In short

Compare tools on one setup. Running 100 prompts on 4 AI apps every day takes 12,000 answers a month. Promptwatch charges $245 for that with 2 seats, and $579 on the plan with 5.
Lumirank runs that setup for $99 a month, on all 6 of its AI apps, with unlimited team members.
Peec AI and Otterly.AI are the closest like-for-like swaps, with unlimited seats, at $330 and $248 a month for the same setup.
Credit-based and weekly tools cost less if you can live with fewer runs: AthenaHQ, Rankscale, and LLMrefs.
The top end moved up. Profound dropped its public self-serve plans in September 2026.

The 10 alternatives at a glance

Prices come from each vendor's pricing page on 24 September 2026, on monthly billing. The third column prices one common setup on each tool: 100 prompts on ChatGPT, Perplexity, Google AI Overviews, and Gemini, run daily.

Tool	Starts at	100 prompts, 4 AI apps, daily	Refresh	Seats
Lumirank	Free, then $29	$99	Daily	Unlimited
Promptwatch (for reference)	$95	$245	Daily	1, 2 or 5
Peec AI	$95	$330	Daily	Unlimited
Otterly.AI	$29	$248	Daily	Unlimited
AthenaHQ	Free credit, then $295	$295, run weekly	You set it	Not stated
Rankscale	$20	$385 daily, $99 weekly	Hourly to monthly	Unlimited
LLMrefs	$79	$79, run weekly	Weekly	Unlimited
Semrush AI Visibility Toolkit	$99	$219, Perplexity weekly only	Daily	Extra users cost extra
Ahrefs Brand Radar	$129 plan	About $330 in checks	Monthly index, custom prompts daily	Per Ahrefs plan
Scrunch	$250	$250, no Gemini	Every 72 hours after day 14	5
Profound	Free trial	Contact sales	Daily	Custom

How Promptwatch works

Promptwatch sends your prompts to the AI apps your buyers use and records what comes back. For ChatGPT, Gemini, Perplexity, Copilot, AI Overviews, and AI Mode, it opens the real apps the way a person would, instead of calling an API. So it records the answer a buyer would actually see.

Each plan limits how many AI answers Promptwatch collects for you in a month. The pricing page has 3 plans: Essential costs $95 a month for 50 prompts, Professional $245 for 150, and Business $579 for 350. Every plan card also says "Track 4 models". That sounds like you can only pick 4 AI apps, but it is really how Promptwatch counts your answers. Essential lets you run 50 prompts on 4 AI apps every day for 30 days, which is 6,000 answers a month. You can spread those answers across any of the 11 AI models Promptwatch supports.

Now take the setup this post uses. Running 100 prompts on 4 AI apps every day needs 400 answers a day, or 12,000 a month. Promptwatch's Professional plan collects up to 18,000 answers a month, so it covers this setup easily.

The number of logins is what pushes the price up. Essential gives you 1 login, Professional 2 and Business 5, and the pricing page offers no way to buy more. So if 3 people on your team need access, you have to buy Business at $579 a month, even though Professional covers your prompts for $245.

Promptwatch also helps you act on what it finds, and that is the main reason people choose it. It works in 4 steps: track the answers, find the gaps, write articles, publish them. Content Gap Analysis compares your pages with the AI answers that leave you out. A Content Agent then writes articles aimed at those answers, 5 a month on Professional and 10 on Business, and can publish them straight to Webflow, Framer, or WordPress. Every plan also includes crawler logs, which show the AI bots that visit your site. The pricing page caps them at 200K, 25M, and 100M on the 3 plans, but does not say 200K of what. API and MCP access come with every plan.

Promptwatch is an Amsterdam company. It launched in April 2025, reached €2M in annual recurring revenue by May 2026, and raised a €6M seed round in July 2026. There are few complaints, and most are about fit rather than quality. The first is price: bigger plans barely lower the cost per prompt. It works out to $1.90 a prompt on Essential, $1.63 on Professional, and $1.65 on Business. One agency user who ran it for 60 days reported more UX errors and bugs than other solutions and no way to escalate. G2 reviewers, as quoted by competitors, mention a steep learning curve. That is thin evidence either way, so weigh it lightly.

Pros: 11 AI models to pick from, collection from the real AI apps, crawler logs, an article writer.
Cons: 1, 2, or 5 seats with no add-on, and price per prompt barely drops as you scale.
Pricing: $95 / $245 / $579 a month for 50 / 150 / 350 prompts.

The 10 alternatives

1. Lumirank

Lumirank covers the monitoring half of the job. It runs your prompts daily on ChatGPT, Gemini, Perplexity, Copilot, Google AI Mode, and Google AI Overviews, in the countries you pick, and tracks visibility, citations, and how you rank against the competitors you add. The Pro plan covers 100 prompts on all 6 AI apps, daily, with unlimited team members, for $99 a month. That is the lowest price on this list for 100 daily prompts on 4 or more AI apps, and Pro comes with a 14-day trial and no card.

It does less than Promptwatch, though. Lumirank does not write articles, read your server logs, or track Claude and Grok, and its pricing page lists the API, MCP server, and exports as coming soon. If Promptwatch's content side is what you use most, it will not replace that. If you need a daily read on 6 AI apps for a whole team, it is the cheapest option here.

Pros: all 6 AI apps on Pro, unlimited team members, daily runs, a free plan with no card.
Cons: no Claude, Grok, or DeepSeek, no crawler logs or article writer, API and exports still coming soon.
Pricing: free for 10 prompts, then $29 / $99 / $399 a month for 25 / 100 / 450 prompts.

2. Peec AI

Peec AI is the closest thing to a like-for-like swap. Its plans cost almost the same as Promptwatch's. The difference that matters for most teams is logins: every Peec plan has unlimited users.

The other difference is which AIs you get. Each plan includes 3 models, picked from ChatGPT, Perplexity, AI Overviews, AI Mode, Gemini, and Copilot, and a 4th costs $85 a month on Pro. So for 100 prompts a day on 4 AI apps, you pay $245 for Pro plus $85 for the 4th app, which is $330 a month. Claude, Grok, DeepSeek, and the rest are Enterprise only, through APIs. One oddity to check during the trial: the pricing table and the model list disagree on whether Perplexity is available below Enterprise.

Peec is a Berlin company that raised a $21M Series A in November 2025 and, according to TechCrunch's sources, passed $10M in annualised revenue by May 2026. This year it added crawl insights for 40+ AI bots, SKU-level AI Shopping tracking in June, and brand fact-checking in August. Pick it if seats are why you are leaving Promptwatch and you don't need the article writer.

Pros: unlimited users on every plan, daily runs, crawl insights and AI Shopping tracking.
Cons: 3 models per plan, a 4th costs $85 a month on Pro, no article writer.
Pricing: $95 / $245 / $495 a month for 50 / 150 / 350 prompts.

3. Otterly.AI

Otterly.AI is the cheapest serious entry point, and every plan has unlimited seats. ChatGPT, AI Overviews, Perplexity, and Copilot are included. AI Mode and Gemini are add-ons at $59 each on Standard, and Claude is $109. To track 100 prompts on ChatGPT, Perplexity, AI Overviews, and Gemini, you need the Standard plan at $189 plus the Gemini add-on at $59, which is $248 a month.

It has grown well past a basic tracker this year. A public API shipped in June and Claude tracking later that month. In August came Agent Analytics, which reads your server logs to show AI bot visits, up to 200K events a month on Standard. Standard and Premium also include MCP access. The GEO Audit checks single URLs for on-page issues. Otterly queries each AI app as a neutral, non-personalized user, daily, per country.

G2 users rate it 4.7 from 54 reviews. The most common complaints are the shift from keyword thinking to prompt thinking, and the sentiment score. It suits you if budget is the reason you are looking, or if a lot of people need to see the dashboard.

Pros: $29 entry plan, unlimited seats, GEO Audit, Agent Analytics from Standard up.
Cons: Gemini, AI Mode, and Claude cost extra, and the sentiment score can be over-sensitive.
Pricing: $29 / $189 / $489 a month for 15 / 100 / 400 prompts.

4. AthenaHQ

AthenaHQ bills in credits, and 1 credit is 1 AI response. Starter covers ChatGPT, Perplexity, AI Overviews, AI Mode, Gemini, Copilot, Claude, Grok, DeepSeek, Meta AI, and Mistral. That is the widest list here at that price, with no add-ons per AI app.

The credits are what limit you. Running 100 prompts on 4 AI apps every day takes 12,000 answers a month, and Starter gives you 3,600 credits. So you have 2 choices: run 30 prompts every day, or run all 100 once a week. Weekly runs use about 1,700 credits a month. You set the schedule yourself, and for a lot of teams weekly is honestly fine.

AthenaHQ went through Y Combinator in early 2025 and raised a $2.2M seed. G2 reviewers give it 4.9 from 46 reviews and name the credit system as the most confusing part. Pick it if you want many AI apps, Claude and Grok included, and can live with weekly runs.

Pros: 11 AI apps including Claude, Grok, and Meta AI with no add-ons, a free plan with $25 of credit.
Cons: credits push a 100-prompt set to weekly runs, and it does not publish how it collects answers.
Pricing: free Essential, then Starter at $295 a month ($245 on annual billing) for 3,600 credits.

5. Rankscale

Rankscale, from Vienna, says it covers 17+ AI apps and models, with no extra fee per AI app. Instead, each AI app costs a different number of credits per run. ChatGPT, Perplexity, Gemini, and AI Mode cost 0.25 credits a run, DeepSeek 1, and Claude 2.

Running 100 prompts on ChatGPT, Perplexity, and Gemini every day takes 9,000 runs a month. At 0.25 credits a run, that uses 2,250 credits, so you need the Growth plan at $385 a month. The calculator shows no separate rate for AI Overviews, so price that 4th AI app during the trial. If you run the prompts once a week instead, the $99 Pro plan is enough. Claude costs the most. Running 100 prompts on Claude alone every day would use 6,000 credits a month, more than Growth includes.

It also does page audits against 200+ factors, prompt research, and shopping tracking. Reviewers on OMR (4.6 from 26 reviews) mention thin docs and limits on client reporting. Agencies and consultants get the most from it: breadth and white-label reports on a small budget.

Pros: 17+ AI apps and models, unlimited seats, page audits, white-label reports and API on Growth.
Cons: each AI app costs different credits (Claude costs 8 times ChatGPT), and docs are thin.
Pricing: from $20, then $99 / $385 / $780 a month for 1,200 / 5,500 / 12,000 credits.

6. LLMrefs

LLMrefs has 1 plan, and next to everything else here it looks like a pricing mistake. The price is low because it checks less often. The FAQ says every keyword is updated at least once per week, and the page frames the price as a limited-time offer.

It is built for people coming from SEO. You give it keywords and it writes the conversational prompts from them, which is a sensible way in if you have never built a prompt set. It adds a query fan-out generator, a crawlability checker, a Reddit threads finder, and an API. Tracking 100 prompts on 4 AI apps once a week fits easily in the $79 plan. There are too few reviews to say much about quality.

Use it as a cheap, wide baseline, or as a second opinion running next to a daily tool.

Pros: 500 prompts on 11 AI apps for $79, unlimited projects and users, prompts written from your keywords.
Cons: weekly refresh, a price framed as limited-time, and no word on how it collects answers.
Pricing: 1 plan at $79 a month.

7. Semrush AI Visibility Toolkit

If you already pay for Semrush, start here. The AI Visibility Toolkit costs $99 a month for 25 prompts, and each extra 50 prompts cost $60. So 100 prompts means 2 extra packs, which is $219 a month for 125 prompts. Semrush One bundles the toolkit with the SEO tools, and its Pro+ plan is $299 a month with 100 prompts.

Coverage is where it gets fiddly. Daily prompt tracking covers ChatGPT Search, AI Mode, AI Overviews, and Gemini. Perplexity shows up only in the Brand Performance report, which refreshes weekly. Semrush says responses are captured from real requests and not via any APIs of LLMs. The AI Search Site Audit flags pages that block AI crawlers like OAI-SearchBot and PerplexityBot, which is worth running on its own.

Semrush is now part of Adobe; that deal closed on 28 April 2026. Extra users cost extra, and Semrush's own pages quote both $45 and $99 a month for one. Pick it if your SEO data already lives in Semrush and 1 login matters more than AI coverage.

Pros: sits inside your Semrush login, audits pages for blocked AI crawlers, a 317M+ prompt database.
Cons: Perplexity only in a weekly report, and extra users and domains cost extra.
Pricing: $99 a month for 25 prompts, plus $60 a month for each 50 more.

8. Ahrefs Brand Radar

Ahrefs Brand Radar works the other way round from Promptwatch. Instead of starting from your prompts, it starts from an index of hundreds of millions of prompts built from People Also Ask questions and Ahrefs' keyword data. The count varies by page: 405M, 451M, and 475M. It shows where any brand appears across AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, and Copilot, collected through the public web interfaces while logged out. That breadth is the point. You find the prompts you would never have thought to track.

To track your own prompts every day, you add custom prompts, and Ahrefs bills them in checks. 1 check is 1 prompt run once on 1 AI app in 1 location. Running 100 prompts on 4 AI apps every day takes 12,000 checks a month. The Lite plan at $129 includes 150 checks, and extra checks cost $100 for 7,000. By the pricing page's numbers, you would pay about $330 a month. Access to the big AI indexes is a separate add-on, and Ahrefs' own pages currently disagree on its price: $199 or $699 a month.

Ahrefs is candid that it does not filter hallucinated links out of its data. Pick it if you already use Ahrefs and want to discover prompts, not only track the ones you wrote.

Pros: an index of 400M+ prompts to find where you appear, Bot Analytics free in beta.
Cons: the chatbot index refreshes monthly, daily tracking is metered, and the add-on price is unclear.
Pricing: needs a paid Ahrefs plan from $129 a month, plus check packs from $50 for 2,500.

9. Scrunch

Scrunch sells 1 self-serve plan. Core covers ChatGPT, Perplexity, AI Overviews, and Copilot for 1 brand in 1 country. Gemini, Claude, AI Mode, Meta AI, and Grok are Enterprise only. It uses a mix of browser automation and official platform APIs. Prompts run daily for 14 days, then every 72 hours by default.

So for 100 daily prompts you pay $250 a month and get 3 of the 4 AI apps in this post's setup, plus 5 logins. Check 1 thing before you sign. Scrunch's help centre says a prompt counts once for each AI app it runs on. By that rule, 125 prompts on 4 AI apps would really be about 31 prompts. The pricing page says "125 unique prompts".

Sitecore announced it was buying Scrunch in June 2026 (Bloomberg reported $225M). Enterprise adds AXP, which serves AI-optimised versions of your pages to AI agents. It makes most sense if you run Sitecore, or care most about how AI agents read your site.

Pros: 5 seats on Core, bot traffic tracking and basic content generation included.
Cons: Core has 4 AI apps and no Gemini, runs every 72 hours after day 14, and prompt counting is unclear.
Pricing: Core at $250 a month for 125 prompts, Enterprise on request.

10. Profound

Profound is the heavyweight. It raised $180M at a $1.8B valuation on 15 September 2026 and says it works with over 1,000 enterprise brands. The same month it took its public Starter ($99) and Growth ($399) plans off the pricing page. What is left is a free 7-day trial on 50 industry prompts across ChatGPT, Gemini, and AI Overviews, and Enterprise. So any setup starts with a sales call.

Profound's strength is its data. It captures answers from the browser. Its Prompt Volumes feature is built on licensed conversations from double-opt-in consumer panels, so it can estimate what people actually ask AI, rather than what they type into Google. Agent Analytics reads server logs from Cloudflare, Fastly, Vercel, and others, and checks which crawlers are genuine.

G2 reviewers rate it 4.5 from 1,128 reviews, the largest review base here by a distance. Pick it if you have an enterprise budget and a team that will act on the findings every week.

Pros: browser capture on up to 9 AI apps, prompt volumes from consumer panel data, crawler analytics.
Cons: no public paid plan, and the free trial runs a fixed set of industry prompts.
Pricing: free 7-day trial, then Enterprise at a custom price.

Also worth a look

A few others did not make the 10 but may fit your case. Evertune asks every prompt 100 times per model and reports share of voice with a stated margin of error, for $800 a month. Its methodology page shows why that matters: 1 run of 1 prompt carries a margin of ±44 points, and 100 runs bring it to ±10. Gauge runs 600 prompts a day on 6 AI apps through logged-out browser sessions for $599 a month, with 5 seats and 18 articles. HubSpot AEO is $50 a month for 25 prompts on ChatGPT, Gemini, and Perplexity, and suggests prompts from your CRM data. Writesonic pairs tracking with a large content tool, but its self-serve plans only track ChatGPT, Gemini, and AI Overviews. Goodie starts at $399 and is the one to look at if you sell products, since it covers shopping assistants like ChatGPT Shopping and Amazon Rufus.

Which one fits your team

Start from the reason you are looking. If it is seats, Lumirank, Peec AI, and Otterly.AI all give unlimited users and run daily, and run 100 daily prompts on 4 AI apps for $99 to $330 a month. If it is budget, Otterly Lite gets you started at $29, and LLMrefs covers 500 prompts for $79 if weekly data is enough.

If you want more AI apps, Claude and Grok included, look at AthenaHQ and Rankscale. Both bill in credits, so decide on your cadence before you compare prices. If your team already lives in an SEO suite, Semrush and Ahrefs save you a login, at the cost of patchy AI coverage (Semrush) or monthly chatbot data (Ahrefs).

At the enterprise end, Profound gives you the deepest data layer. And if Promptwatch's Content Agent is the part you use most, the cheaper trackers here will not replace it. Of the 10, Profound writes articles and Scrunch Core includes basic content generation. Outside the 10, Evertune, Gauge, and Writesonic do both.

Before you switch, run both for 2 weeks

Scores do not compare across tools out of the box. Each vendor defines visibility a little differently, collects from different places, and matches brand names its own way. So compare what each tool records for the same prompts, rather than their headline scores.

Take 20 of the prompts you track. Mix category questions with head-to-head ones against your main rivals.
Load the same 20 into Promptwatch and the candidate, on the same AI apps and the same country.
Run both for 14 days. Compare the mention rate per AI app over the whole window, not day by day, because single runs vary a lot.
Each week, run 5 of the prompts yourself in a logged-out browser from the country you target, and see which tool's record matches what you saw.
Then price your full job, seats included, on whichever tool matched best.

Reference

10 Best Promptwatch Alternatives in 2026

Profound Is Pivoting: 10 Tools That Still Publish Their Prices

Saptarshi Paul — Thu, 01 Oct 2026 05:48:49 +0000

On 15 September 2026, Profound announced $180M in new funding at a $1.8B valuation. Days later, the Starter and Growth plans, $99 and $399 a month, disappeared from its pricing page. If your team was on Growth, you now need a replacement.

Profound tracks how often AI apps name your brand. It does four jobs: it runs your prompts daily and records answers, it scores your visibility, it turns gaps into content with an AI Marketer, and it watches AI bot traffic with Agent Analytics. No tool in this post does all four. But for daily prompt tracking, several tools are cheaper and more transparent.

This post covers 10 alternatives that publish prices, bill monthly, and can start the same day. The focus is on how each meters usage and where each fits Profound's workflow.

What Profound's four parts map to

Profound's four parts are:

Answer Engine Insights: runs prompts daily and records answers.
Visibility Score: share of answers that name you.
AI Marketer and Agents: turns visibility gaps into content.
Agent Analytics: tracks crawler and bot behavior.

The table below shows the closest match in this post for each part.

Profound part	Closest match in this post
Answer Engine Insights	All 10. Widest coverage: Rankscale (17+ AI apps), AthenaHQ and Promptwatch (11 each)
Visibility Score	Lumirank, Otterly.AI, Peec AI
AI Marketer and Agents	Promptwatch Content Agent, basic content generation on Scrunch Core
Agent Analytics	Peec AI, Ahrefs bot crawler logs

That last row is where alternatives are cheapest and most alike. The choice comes down to how each tool meters usage.

Visibility Score and the API gap

Profound's case for tracking the public app rather than the API is that 'API responses and front-end responses from the same model can diverge', because the public app carries context an API call does not see.

Profound's Visibility Score is the share of answers that name you, out of the answers that name at least one brand. If ChatGPT often replies with general advice and no vendor names, Profound's score runs higher. A 15-point drop after a switch may be nothing more than a new denominator. The same answers can produce two very different numbers depending on the definition.

Profound publishes no Enterprise price. The $2,000 to $5,000 or more a month that third-party reviews quote comes from user reports, not from Profound, so treat it as a rough guide. Profound licenses conversations from multiple, double-opt-in consumer panels, covering hundreds of millions of prompts a month in 10 countries, updated weekly.

The AI Marketer runs Agents that turn visibility gaps into content, and each Agent run spends credits. That is a separate cost from prompt tracking.

How the alternatives meter usage

You pay for questions. Each question runs on every AI app the plan covers.

Lumirank works this way from Pro up, as do LLMrefs and Semrush's toolkit. Adding an AI app costs nothing within the plan, so the bill is predictable.

Otterly.AI includes four apps and charges $59 a month each for Gemini and AI Mode on Standard. Peec AI includes three models and charges $85 for a fourth on Pro. If the payroll team needs Gemini, Otterly.AI's Standard plan costs $248, not $189.

Credits are the most flexible unit and the hardest to forecast. Ahrefs and Semrush use credits where each AI app costs a different rate. Claude can cost eight times ChatGPT.

Plan limits differ:

Lumirank, Otterly.AI, Peec AI, Rankscale and LLMrefs have unlimited users.
Promptwatch has 1, 2 or 5 seats.
Scrunch has 5 seats.
Semrush charges per extra user.

Lumirank and AthenaHQ have free plans.

Plans with 100 to 150 prompts cost $99 to $250 a month across Lumirank, Otterly.AI, Semrush, Peec AI, Promptwatch and Scrunch. Every one lists a lower monthly price than Profound's old Growth plan at $399 a month paid yearly.

The 10 alternatives

Lumirank

Lumirank runs each prompt once a day on every AI model in your plan, in the countries you pick. A prompt is 'one question in one country'. The same question tracked in the US and Canada uses two prompts.

Pro is $99 for 100 prompts on all six apps: ChatGPT, Gemini, Perplexity, Copilot, Google AI Mode and AI Overviews. Scale is $399 for 450 prompts. Extra countries use extra prompts, but nothing is charged per AI app or per seat. Every plan has unlimited team members. The free plan covers 10 prompts on two apps, and Pro has a 14-day trial with no card.

Compared with Profound's old Growth plan, Pro has the same prompt count on twice the AI apps, Gemini included, for a quarter of the price, billed monthly.

Lumirank's home page shows daily tracking across six AI apps. Screenshot: lumirank.ai, September 2026.

What it lacks: no prompt volume data, no crawler logs, no content agent. It does not track Claude, Grok or DeepSeek. Its pricing page lists the API, MCP server and exports as 'coming soon'. That matters if Profound's data fed a BI tool. It also does not say whether answers come from web interfaces or APIs.

Pros: all six AI apps on Pro, unlimited team members, daily runs in multiple countries, monthly billing.
Cons: no Claude, Grok or DeepSeek, no prompt volumes, no crawler logs, no content agent, and API and exports are still 'coming soon'.

Otterly.AI

Otterly.AI queries each AI app daily, per country, as a neutral, non-personalized user. On Standard, AI Mode and Gemini cost $59 a month each and Claude costs $109. Standard with all three added comes to $416.

Crawler data sits on the same $189 plan as prompt tracking. A public API arrived in June and Claude tracking arrived later that month. Standard and Premium include MCP access. The GEO Audit checks single pages for on-page issues.

Otterly.AI includes four apps and charges extra for Gemini and AI Mode on Standard. The list price holds only if the included apps are the ones your buyers use.

Peec AI

Peec AI meters prompts and models. Each plan includes three models chosen from ChatGPT, Perplexity, AI Overviews, AI Mode, Gemini and Copilot. A fourth model costs $85 a month on Pro. Peec also covers part of Agent Analytics with crawl insights for 40+ AI bots.

Essential is $95 for 50 prompts, Professional is $245 for 150, and Business is $579 for 350. Seats are metered as well: 1, 2 and 5. Headcount can push you up a plan before your prompt count does.

Peec AI pitches itself as AI search analytics for marketing teams. Screenshot: peec.ai, September 2026.

Cons: three models per plan, a fourth costs $85 a month on Pro, and no content agent or prompt volume data.

LLMrefs

LLMrefs has one plan at $79 a month: 500 prompts on 11 AI apps, with unlimited projects and users. That is about 16 cents a prompt, the lowest on this list by a distance. The FAQ says every keyword updates at least once per week, and the price is framed as limited-time.

You give it a keyword and it writes the conversational prompts. That is a gentle way in if Profound's suggested prompts were your starting point.

LLMrefs starts from a keyword, not a prompt. Screenshot: llmrefs.com, September 2026.

Also worth a look

The remaining six tools compete more directly with Profound and cost more.

Ahrefs runs prompts through the public web interfaces of AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini and Copilot, logged out. Its index covers 400M+ prompts. A 317M+ prompt database is a strong asset. But Ahrefs does not filter hallucinated links out of its data. Access to big AI indexes is an add-on, and its own pages disagree on the price: $199 or $699 a month.
Semrush One bundles the toolkit with SEO tools. Its Pro+ plan is $299 a month for 100 prompts. It includes bot traffic tracking and basic content generation.
AthenaHQ supports 11 AI apps including Claude and DeepSeek. It has a free plan with $25 of credit.
Rankscale covers 17+ AI apps. Credits are the catch because every app and run spends them.
Scrunch Core costs $250 a month for 125 prompts and 5 seats. It includes Copilot, bot traffic tracking and basic content generation. It runs every 72 hours after day 14, and there is no Gemini on Core.
Promptwatch is the only one of the 10 with a full article writer. It tracks 11 AI apps and has 1, 2 or 5 seats. Its pricing page lists no add-on.

How to compare a candidate against Profound

If you are a current customer with a renewal due, put Profound next to one candidate. If not, put two candidates next to each other.

Use this test:

Write 25 prompts: 15 category questions such as 'best payroll software for a 50-person US startup', and 10 that compare you with your two closest rivals.
Load all 25 into both tools, on the same AI apps, with the US as the country.
Each week, ask 5 of the prompts yourself in a logged-out browser from the US.
Compare the lists. If they mostly agree, both tools are reading the same answers.

Then price your full prompt list, AI apps and seats on the tool you keep. If the lists diverge, check which tool explains its collection method. If panel data on AI prompts matters to you, or your procurement team needs SSO and SOC 2, Profound's Enterprise plan is still worth the call.

Reference

10 Best Profound Alternatives in 2026

Cloud Computers for AI Agents in 2026: A Practical Comparison

Saptarshi Paul — Thu, 01 Oct 2026 04:29:47 +0000

An AI agent that only talks is cheap to host. The moment it needs to install packages, run tests, start a server, or leave a job running overnight, it needs a computer of its own. Handing it yours means giving a language model a shell on the machine that holds your SSH keys.

A whole category of products now rents that computer by the second. They look alike on their landing pages, but underneath they make very different choices. machine0 gives you a full KVM virtual machine that runs until you stop it. OpenComputer caps every sandbox at 8 hours, hibernation included. E2B restores a Firecracker microVM from a memory snapshot in about 150 ms. Daytona runs plain containers unless you ask for a VM. I went through the docs, pricing pages, and SDKs of seven of them, installed each SDK, and ran the examples below up to the API-key check. All facts and prices for those seven are as of September 26, 2026. Five newer ones get a short honorable mention near the end.

The running example throughout is one ordinary agent job: clone a Node.js repo, run npm test, start a dev server on port 3000, and show someone the result.

Summary

Pick by lifetime first. machine0 and Fly.io Sprites give an agent a machine that persists indefinitely. OpenComputer v2 sandboxes end after a hard 8 hours. E2B, Vercel Sandbox, and Modal cap a continuous session at 24 hours on paid plans.

Isolation is not uniform. machine0 and OpenComputer run KVM virtual machines, E2B, Vercel, and Sprites use Firecracker microVMs, Modal uses gVisor, and Daytona defaults to Linux containers.

Of these seven, only E2B pauses every sandbox with memory intact and keeps the paused sandbox with no expiry. Daytona does it on VM sandboxes only; the rest save the filesystem.

For a 2 vCPU / 4 GB machine running flat out for an hour, machine0 costs $0.052, E2B and Daytona $0.166, Modal $0.238, and OpenComputer $0.378. machine0 bills while stopped, though, and the usage-billed ones cost far less when the agent is mostly waiting on a model.

Need GPUs? machine0, Daytona, and Modal have them. Need a desktop for computer use? E2B Desktop, Daytona Computer Use, or Orgo.

What a cloud computer for an agent actually is

Every product here does the same four jobs. It boots an isolated Linux machine, gives the agent a way to run commands in it (an SDK call, SSH, or an MCP tool), exposes a port so a human can see the result, and decides what survives when the machine stops. The interesting differences are in how each job is done.

Isolation comes in three strengths. A full virtual machine under KVM gets its own kernel and looks like a normal server, so Docker, kernel modules, and GPUs work. A Firecracker microVM is a stripped-down KVM guest built to start fast; E2B's runtime README explains that "creating" a sandbox means "restoring one, not booting a kernel", with memory pages loaded lazily on first access. Containers share the host kernel and rely on namespaces, and gVisor sits in between by intercepting system calls in a user-space kernel.

Persistence is where the category splits into two camps. Persistent computers keep a disk and run until you stop them, like a VPS an agent can drive. Sandboxes are built to be created per task, used for minutes, and thrown away, with snapshots as the way back. Neither is better. A coding agent that works on one repo for a week wants the first kind. A product that runs untrusted code for thousands of users wants the second.

Where each product puts the boundary between the agent's code and the host.

Comparison table

The price column is the list rate for 2 vCPU and 4 GB running at full CPU for one hour. Where a product bills on actual usage, the real bill for an agent that spends most of its time waiting on model calls is lower.

Data: machine0, E2B, Daytona, Modal, Fly.io, Vercel, and OpenComputer pricing pages, September 26, 2026.

Product	Isolation	Max lifetime	Keeps memory on pause	GPU	2 vCPU / 4 GB, 1 h
machine0	Full KVM VM	None, runs until stopped	No (suspend is disk snapshot)	Up to 8x H200	$0.052
OpenComputer (v2)	Full KVM VM	8 h, hard	No	No	$0.378
E2B	Firecracker microVM	1 h Hobby / 24 h Pro per session, resets on pause	Yes, kept indefinitely	No	$0.166
Daytona	Container (VM optional)	None, auto-stops after 15 min idle by default	VM sandboxes only	Up to 8 GPUs	$0.166
Fly.io Sprites	Firecracker microVM	None, sleeps when idle	While warm only	No	$0.315 (usage-billed)
Vercel Sandbox	Firecracker microVM	45 min Hobby / 24 h Pro per session	No (filesystem snapshot)	No	$0.341 (active CPU)
Modal Sandboxes	gVisor (VM in beta)	24 h	Alpha only	T4 to B300	$0.238

machine0: a real VM your agent keeps

machine0 launched on Hacker News on August 18, 2026 as a YC Summer 2026 company, and it is the most traditional product here. The founder's launch post puts it plainly: "Every machine is a full KVM virtual machine, not a container or sandbox." Its FAQ says the VMs run on DigitalOcean infrastructure across five regions. You get a dedicated public IP (kept across stop, replaced on suspend), SSH as user ubuntu with password login disabled, and an authenticated HTTPS URL at https://<name>.mac0.io that proxies to port 80.

There is no language SDK. The interface is a CLI (list and detail commands take --json) plus a remote MCP server, which is a sensible choice when the client is an agent that already knows how to use a shell. The worked example looks like this with CLI version 1.0.164:

npm install -g @machine0/cli        # or: curl -LsSf https://machine0.io/install.sh | sh

machine0 new agent-box --size large --region eu --image ubuntu-24-04-loaded

machine0 ssh agent-box "git clone https://github.com/your-org/app.git && cd app && npm ci && npm test"

machine0 suspend agent-box           # snapshot the disk, delete the instance, stop compute billing

machine0 start agent-box             # restore later, with a new IP

The loaded image ships Docker, Node.js, Python, Go, Rust, Bun, Claude Code, Codex, and OpenCode. To give Claude Code control of your fleet, run claude mcp add --transport http machine0 https://app.machine0.io/mcp.

The tradeoffs are the ones you would expect from a VPS. Provisioning takes one to three minutes, not milliseconds. The docs say a stopped VM is "still billed at full rate", so you have to suspend it to save money, and suspend keeps the disk but not the running processes. Your dev server on port 3000 is not on the HTTPS URL until you move it to port 80 or open the port with ufw. In exchange, sizes go from small (1 vCPU, 1 GB, $0.013/hour) to 6xl (60 vCPU, 240 GB) and GPU sizes up to 8x H200 at $39.336/hour, billed per minute.

OpenComputer: SDK-first VMs with an 8-hour ceiling

OpenComputer comes from the team behind Digger and is open source under Apache-2.0 (554 stars). Its docs describe a sandbox as "Not a container. Real VM with its own kernel, memory, and disk", and the self-hosting guide describes a control plane driving QEMU/KVM on KVM-capable Linux workers.

If you read about OpenComputer earlier this year, much of it has changed. The v1 runtime is retired, and a 0.x SDK now gets 410 Gone. On v2, the lifetime page is blunt: "The ceiling is 8 hours, and it counts running and hibernated time together. It cannot be extended." Checkpoints no longer capture memory, forking from a checkpoint is not available yet, RAM comes in fixed steps of 1, 2, 4, or 8 GB, and disk is fixed at about 16 GB. The homepage itself now leads with a separate product for deploying agents as TypeScript functions.

The sandbox SDK is still the cleanest API of the lot. This is the docs example, which I ran against @opencomputer/sdk 2.1.2 up to the 401 for a missing key:

import { Sandbox } from "@opencomputer/sdk";

const sandbox = await Sandbox.create();

const result = await sandbox.exec.run("echo Hello from $(uname -a)");

console.log(result.stdout);

await sandbox.kill();

sandbox.getPreviewDomain(3000) returns a public URL for the dev server; preview URLs are public by default. Pricing is per second: $0.0063 per minute for 4 GB and 2 vCPU, with $10 of free credit. Use it for bounded jobs that finish inside a working day.

E2B: Firecracker microVMs restored from snapshots

E2B is the sandbox most agent frameworks reach for first. Its SDK repo has 14,008 stars and its backend (e2b-dev/runtime) is Apache-2.0 and self-hostable. Because a new sandbox is a snapshot restore, the docs put startup at about 150 ms.

The feature that sets it apart is pause. The persistence docs say pausing saves "not only state of the sandbox's filesystem but also the sandbox's memory", that a paused sandbox "is kept indefinitely", and that resume takes about a second (pausing takes about 4 seconds per GiB of RAM). Billing stops while paused. Tested against e2b 2.51.0:

from e2b import Sandbox

# Auto-pause instead of killing when the 10-minute timeout hits
sandbox = Sandbox.create(timeout=600, lifecycle={"on_timeout": "pause"})
result = sandbox.commands.run("git clone https://github.com/your-org/app.git app && cd app && npm ci && npm test")
print(result.stdout)
print(sandbox.get_host(3000))  # public URL for a dev server on port 3000

sandbox_id = sandbox.sandbox_id
sandbox.pause()                        # filesystem + memory saved
sandbox = Sandbox.connect(sandbox_id)  # resumes where it stopped

The limits: the default sandbox is 2 vCPU and 512 MiB of RAM, and you change that by building a template. The FAQ is explicit that sandboxes are "CPU-only". SSH needs a custom template with a WebSocket proxy. Hobby is free with $100 of one-time credit; Pro is $150/month.

Daytona: containers by default, VMs and GPUs on request

Daytona charges exactly E2B's per-second rates ($0.0504 per vCPU-hour, $0.0162 per GiB-hour) and adds $200 of free compute. It also covers more ground: GPU sandboxes up to 8 GPUs, Windows VMs, built-in SSH (daytona ssh), an MCP server in the CLI, and SDKs for Python, TypeScript, Ruby, Go, and Java.

The catch is near the top of its sandbox docs: "Sandboxes run as Linux containers by default." Memory-preserving pause and fork only work on VM sandboxes. The default auto-stop is 15 minutes, and the docs warn that "Merely having a script or background task running is not sufficient to keep the sandbox alive", so a long npm test with no API calls can be stopped mid-run unless you set auto_stop_interval=0. Also note that the 71,697-star daytonaio/daytona repo stopped receiving updates in June 2026, when core development moved to a private codebase.

from daytona import Daytona, DaytonaConfig

daytona = Daytona(DaytonaConfig(api_key="YOUR_API_KEY"))
sandbox = daytona.create()
response = sandbox.process.exec("echo 'Hello from Daytona'")
print(response.result)

That ran on daytona 0.218.0 until the API rejected the placeholder key.

Fly.io Sprites: persistent microVMs that sleep

Sprites sit between machine0 and a sandbox. Each one is a Firecracker microVM with 8 vCPUs, 100 GB of storage on Ubuntu 25.10, and its own HTTPS URL. When idle it sleeps, and the next command or HTTP request wakes it: about 100 to 500 ms from a warm suspend with processes intact, or 1 to 2 seconds from cold, which keeps the disk but loses memory. Billing covers only the CPU and memory actually used ($0.07 per CPU-hour, $0.04375 per GB-hour), with $30 of trial credit.

curl -fsSL https://sprites.dev/install.sh | sh
sprite create agent-box
sprite exec -- echo "Hello, Sprites!"
sprite checkpoint create --comment "tests passing"

Checkpoints are copy-on-write disk snapshots. The URL routes to port 8080, not 3000, and needs a token unless you run sprite config update --url-auth public.

Vercel Sandbox and Modal

Vercel Sandbox says "each sandbox runs in its own Firecracker microVM with a dedicated kernel", snapshots the filesystem when a sandbox stops, and bills only active CPU: "Time spent waiting for I/O (such as network requests, database queries, or AI model calls) does not count." That matters for agents, which spend most of their time waiting on a model. Hobby is free with 5 CPU-hours a month and 45-minute sessions. The SDK is @vercel/sandbox (3.5.0), and it authenticates through a Vercel project's OIDC token, so it fits best if you already deploy there.

Modal Sandboxes use gVisor, with full-VM sandboxes in beta. They bring GPUs from T4 to B300 and scale to very large numbers of concurrent sandboxes, but the maximum lifetime is 24 hours and there is no pause, only filesystem snapshots (memory snapshots are alpha and expire after 7 days).

If your agent needs a desktop

Computer-use agents that click through a GUI need a screen, not just a shell. E2B Desktop (pip install e2b-desktop) gives an Ubuntu 22.04 XFCE desktop with VNC streaming. Daytona's sandbox.computer_use.start() launches Xvfb, XFCE, and noVNC on Linux or Windows. Orgo sells always-on desktops from $29/month with an MCP server exposing 51 tools. Cua, in the honorable mentions below, rents Linux and Windows desktop pools too. Avoid older guides that point at Cloudflare's sandbox desktop or Scrapybara: Cloudflare removed the desktop in Sandbox SDK 0.12.0, and Scrapybara shut down on October 15, 2025. For the agent loop itself, see our guide on how AI agents control a computer.

Honorable mentions

These five are newer or a slightly different shape. I read their docs and pricing pages on September 29, 2026, but did not run them.

Boat (YC Fall 2026) rents full Ubuntu VMs with /dev/kvm passed through, at the lowest price in this post: $0.018/hour for 2 vCPU / 4 GB, billed per second on top of a $20/month plan. The catches: the FAQ says vCPUs are shared, it is hosted in the EU only, and snapshots keep the filesystem but not memory.

Freestyle (YC S24) runs full VMs on its own bare-metal racks that pause with memory intact, and new VMs can start from memory-and-disk snapshots. 2 vCPU / 4 GiB is about $0.13/hour, and the monthly allowance covers roughly the first 100 hours. The SDK is TypeScript only, and its own docs say it oversubscribes hosts, so skip it for CPU-heavy CI.

Cua (YC X25) is best known for its open-source computer-use tooling (trycua/cua, 27,148 stars), and it also rents pools of Ubuntu 24.04 and Windows Server 2022 desktop VMs at about $0.18/hour for 2 vCPU / 4 GiB. Warm pools keep billing while idle, and snapshots are not available on its hosted fleets yet.

InstaVM runs Firecracker microVMs at E2B's exact rates ($0.166/hour for 2 vCPU / 4 GB), with SSH, private share URLs, and a noVNC desktop for computer use. Sessions are ephemeral by default and last up to 24 hours, and suspend with memory is still behind a server-side feature flag.

Prized (YC S26) is not a sandbox API. It is a persistent EC2 dev box that syncs both ways with your laptop, so your own coding agents keep running when the lid closes. It is private by design ("A box cannot open a port to the internet"), costs $25/month for 2 vCPU / 4 GB, runs in US West only, and is very new: it pivoted to this in August 2026.

Letting a cloud agent reach your laptop with Pinggy

Every product above can publish a port from the cloud machine. The reverse direction is the one that trips people up: the agent is in the cloud, but the API it has to test against, a local model server, or a webhook receiver is running on your laptop behind NAT. A Pinggy tunnel fixes that with one command on your laptop and nothing installed:

ssh -p 443 -R0:localhost:8000 free.pinggy.io

It prints an https:// URL that forwards to localhost:8000. Pass it to the agent as an environment variable, for example with machine0 env set on the profile you create VMs from, or Sandbox.create(envs=...) in E2B. Free tunnels last 60 minutes; for an agent that runs all day, use a Pro token (<token>@pro.pinggy.io) to get a persistent URL. If the service has no auth of its own, add some, because the URL is public. Our n8n and Pinggy guide walks through a longer version of this setup.

How to choose

Start with how long the work runs. If an agent keeps one environment for days, with a repo, caches, and a running service, pick a persistent computer: machine0 if you want a plain VM, GPUs, or a low hourly rate, and Sprites if you want it to sleep and cost nothing when idle. If each task is short and you create machines from code, pick a sandbox. E2B is the safest default because it gives every sandbox a microVM and keeps memory on pause. Daytona suits workloads that need GPUs, Windows, or SSH, and Vercel suits teams already deploying there. Use OpenComputer for jobs that finish inside 8 hours, where you want an open-source platform you can self-host.

Then check the limit you are most likely to hit: an idle timeout that kills a quiet npm test, a default of 512 MiB of RAM, a stopped VM that still bills, or a preview URL that is public by default.

Conclusion

Every product in the comparison table has a free credit or a free tier except machine0, which starts from a $5 prepaid top-up, so the cheapest way to decide is to run your own agent's workload on two of them. Time how long the first command takes, leave the machine idle for 20 minutes and see whether it is still there, and compare the bill after a day. Those three checks will tell you more than any table, including this one.

Instantly share your localhost over secure public URLs - no signup, no config.

Reference

Best Cloud Computers for AI Agents in 2026: machine0 vs OpenComputer vs E2B, Daytona and Sprites