DEV Community: VOXA

Best AI design tools for people who are not designers

VOXA — Fri, 06 Nov 2026 17:38:35 +0000

The short answer

Five tools cover almost every design job a non-designer runs into: Canva for general layouts, Photoroom for product photos, Ideogram for images with readable text, Recraft for vector files that stay sharp at any size, and Figma for anything that has to be handed to a developer. Each one is good at one job and mediocre at the rest. The mistake is trying to do all five jobs in one tool.

Pick the job first, then the tool

"AI design tool" is not a category, it is a shelf. The job underneath decides everything: a social post is a layout problem, a product photo is a lighting and masking problem, a logo is a vector problem, and a landing page is a handoff problem. Tools that try to cover all four end up doing each one at 70 per cent, which is fine for a quick post and painful for anything that ships.

So work backwards. Write down the artefact you need, the format it has to be in, and who touches it after you. If the answer involves a printer, a developer, or a brand team, the tool choice narrows immediately.

The shortlist, by job

Canva (Magic Studio). The default for anything with a template: social posts, decks, one-page PDFs, simple video. The AI features generate layouts, resize across formats and remove backgrounds. Best when you need something decent in twenty minutes. Weakest at original visual identity, because everything it produces is anchored to the templates everyone else uses.
Photoroom. Product photos. Drop the phone shot in, get a clean cutout and a studio background. This is the tool that most replaces a photographer for small ecommerce, and the free tier handles occasional use. Watch the edges on hair, glass and transparent objects.
Ideogram. Images that contain words. Most image models still mangle text; this one renders short phrases and signs more reliably than the rest. Good for thumbnails, posters and mock packaging where a word has to be legible.
Recraft. Vector output. If the file has to be a logo, an icon set or anything printed large, you need SVG or EPS, not a 1024-pixel JPEG. Recraft generates vector natively, which saves a redraw in Illustrator later.
Figma. Not really an AI design tool, and that is the point. It is where a design stops being an image and becomes a file with layers, spacing and states that a developer can build from. Its AI features are conveniences on top of that structure.

Where they actually fail

Three failures show up again and again, and knowing them saves a wasted afternoon.

Text inside images. Any model that draws letters will eventually draw a letter that is almost right. At small sizes it looks fine; printed at A2 it is a typo. If the text matters, set it as text, not as pixels.

Brand consistency. Generated output drifts. Ten posts made in ten sessions will not look like one brand unless you fix colours, type and layout rules in a template and only vary the photo. The template is the product; the generation is the raw material.

Print and file formats. A JPEG that looks sharp on a laptop can be unusable on a 3-metre banner. Anything physical needs vector or a very high-resolution raster, and most AI tools output the latter at best. For embroidery, signage and packaging, plan on a real vector step.

How to judge output in thirty seconds

Zoom to 200 per cent and look at edges. Generated images often have soft, smeared boundaries where objects meet, especially at the corners of shapes. Then look at hands, reflections and shadows, which are where the model's internal physics breaks. Finally, check whether the image makes sense at thumbnail size, because that is how most people will see it.

For layout tools, the test is different: does the design survive a text change? Double the headline and delete a paragraph. If the layout collapses, the template was doing the work and you were not.

What to pay for

Start free and stay free until a specific limit stops you. The usual limits are export resolution, watermark removal, brand kits and seats. Paying for a seat before you have a repeatable workflow is the common mistake: you end up paying monthly for a tool you open twice.

If you are choosing one paid subscription and you make content daily, pick the layout tool, because that is where the time goes. If you sell physical products, pick the photo tool, because a clean product image converts better than a prettier post.

The workflow that actually holds up

Generate raw material, then assemble by hand in a template. In practice: product photo in Photoroom, background or illustration in Ideogram or Recraft, layout and text in Canva, and anything that has to be built handed to Figma. That keeps generation where it is strong and keeps the parts that need consistency under human control.

One rule ties it together: never let the tool own the file format you cannot open elsewhere. Export your brand assets, keep the source files, and assume you will change tools within two years. The design survives; the subscription does not have to.

Originally published at voxahq.in.

Japan put AI inside an emergency cybersecurity package

VOXA — Fri, 06 Nov 2026 17:37:20 +0000

The short answer

Japan's government announced an emergency cybersecurity package on 6 November, after a run of cyberattacks that leaked personal data and caused financial losses. The package creates a single contact point for companies, pushes AI into the job of finding vulnerabilities, and puts money into protecting hospitals and training security staff. Prime Minister Sanae Takaichi ordered it on 26 October, and it comes on top of an October legal change that lets police and the Self-Defence Forces reach into attackers' systems.

What is in the package

The measures reported so far fall into four groups. First, a cross-ministerial system to analyse damage and handle consultations, with a unified point of contact so a company under attack does not have to work out which agency to call. Second, stronger cybersecurity for medical institutions, which have been a repeated weak point because they run old systems and hold sensitive records. Third, recruitment and training of security personnel, which is the bottleneck in every country. Fourth, promoting the use of AI to help companies identify and fix their own vulnerabilities.

The framing matters. This is not a bill with new penalties. It is an operational package: who answers the phone, who analyses the incident, and what tools companies are encouraged to use before an incident happens. That is often slower to write than legislation and faster to actually deploy.

The package follows a first ministerial panel on cybersecurity on 26 October, where the prime minister instructed ministers to put it together. Two weeks from instruction to announcement is quick by Japanese policy standards, which tells you the incidents behind it landed hard.

Why now

The stated trigger is a series of attacks that produced personal information leaks and financial losses. Japan has had a rough run on this front: supply chain compromises at vendors that serve hundreds of companies, ransomware against hospitals and local government, and credential theft that turns into payment fraud. Each incident is small on its own and expensive in aggregate.

The structural problem is the same one every mid-sized economy has. Most companies do not have a security team. They have an IT manager, an outsourced provider, and a firewall. When something goes wrong, the first hour is spent working out who to tell. A unified contact point sounds bureaucratic, and it is also the single cheapest thing a government can build.

The AI part, and what it can and cannot do

Using AI to identify vulnerabilities is the part that will get the headlines, and it is worth being precise about it. There are two different jobs here. One is scanning code and configuration for known classes of weakness: missing patches, exposed services, weak authentication, misconfigured storage. Models are decent at this, and they scale in a way human reviewers do not. The other is finding genuinely novel attack paths in a complex system, which is much harder and where confident wrong answers are expensive.

The realistic outcome is that AI raises the floor. More small companies get a basic review they could never afford to buy. The ceiling, meaning the deep adversarial work, stays with people. A package that funds both, tools for the many and trained staff for the few, is better designed than one that only buys software.

There is also a symmetry nobody enjoys mentioning. The same models that help defenders find weaknesses help attackers find them. The gap between the two sides is not capability, it is process: who patches first, who has logs, who notices.

Active cyber defence is the harder half

Japan revised related laws in October to allow police and the Self-Defence Forces to access attackers' computers in order to disable them. That is active cyber defence, and it is a significant step for a country with a pacifist constitutional tradition. The package is expected to include measures to staff that capability.

The operational questions are unresolved everywhere this has been tried. What counts as an attacker's system when infrastructure is shared? What happens when the trail runs through a third country? Who authorises the action, and how quickly, given that most intrusions are over in hours? Building the legal authority is the easy part. Building a unit that can act inside the window is the hard part, and it takes years.

For companies, the practical read is that offensive capability will not protect them. It changes the risk for attackers at a national level. It does not patch a hospital's server.

What it means outside Japan

Three things are worth copying. The single contact point, because it costs little and removes the worst hour of an incident. The focus on medical institutions, because they are consistently the weakest link and the most damaging to lose. And the split between tooling for everyone and trained people for the few, because funding only one of those produces a false sense of coverage.

What is harder to copy is the speed. Two weeks from a prime ministerial instruction to a published package requires a civil service that already has the analysis done. Most countries produce the analysis after the incident and the package a year later.

What to watch

Whether the unified contact point gets a budget line and staff, or stays a phone number that routes to a queue.
Whether the AI vulnerability tools are procured centrally or left to companies to buy, which decides whether small firms actually get them.
Whether the medical institution funding comes with mandatory patching, because grants without requirements tend to buy software nobody deploys.
Whether any of the incidents behind the package are disclosed in detail. Packages built on unnamed breaches are hard to audit later.

For a small company anywhere, the lesson is not to wait for a national package. The three moves that prevent most of these incidents are unglamorous: turn on two-factor authentication everywhere, remove accounts belonging to people who left, and test a restore rather than assuming the backup works. Every one of those is free, and every one of them is still skipped.

Originally published at voxahq.in.

Canva cut 1,200 jobs and said small changes will not do

VOXA — Fri, 06 Nov 2026 17:35:52 +0000

The short answer

Canva is cutting about 22 per cent of its workforce, roughly 1,200 people, either through redundancy or relocation. Chief executive Melanie Perkins told staff that "small changes won't be enough" as the company rebuilds around AI. It is the largest restructuring in the Australian design company's history, and it lands at a moment when the tools Canva sells are being rebuilt by the same technology it is deploying.

What was announced

The cuts were communicated to staff this week. Canva has not published a country-by-country breakdown of who is affected, and it has not said how many of the 1,200 roles are being removed versus moved to another location. The company has framed the change as a structural one rather than a cost exercise: a smaller organisation, organised differently, built around AI-assisted work.

Perkins's language in the internal note was unusually direct for a company that has spent a decade growing headcount. She wrote that she had reached the conclusion that small changes would not be enough. That is a different sentence from the usual "restructuring to focus on our priorities", and it points at a reorganisation rather than a trimming.

The timing is also notable. Canva is private, profitable for years, and has a balance sheet that did not force this decision. It already cut 32 roles in its finance division earlier this year, a round that was never announced publicly. A second, much larger round eight months later suggests the first was a test of a thesis rather than an isolated correction.

The numbers that matter

Twenty-two per cent is the headline, and it is high for a company that is not in financial distress. For context, technology companies that cut in 2026 have mostly done so in single digits, usually framed around a specific division. A cut of this size across the whole company means entire layers of middle management, support functions and specialist teams are being reworked at once.

One thousand two hundred people is also large in absolute terms for Australia, where Canva is one of the few globally competitive software companies. The country's tech sector is small enough that a cut of this size is felt across the local hiring market, not just inside one company.

What is missing from the announcement matters as much as what is in it. There is no published productivity number, no statement about how much faster a team now ships, and no breakdown of which functions are shrinking. As with every AI-attributed cut this year, the claim rests on management's account of its own reasoning rather than a measurement anyone outside can check.

Why design tools are exposed

Design software sits in an uncomfortable place in this cycle. Its users are not engineers who need a compiler, but people producing a specific artefact: a post, a deck, a thumbnail, a product photo. That artefact is exactly what image and layout models have become good at generating. The work does not disappear, but the number of steps between an idea and a finished file drops sharply, and every step removed was somebody's job.

Canva's own product strategy has been to absorb that shift rather than resist it. Magic Studio, its AI feature set, generates layouts, writes copy, removes backgrounds and resizes designs across formats. That is a sensible product decision and an awkward staffing one: the same features reduce the amount of manual production work per design, which reduces how many people you need per design team.

The competitive pressure is coming from two directions at once. Upstarts ship narrow AI design tools that do one job well, and the large platform players bundle similar features into products that customers already pay for. A company selling a horizontal design tool has to fund AI development against both, from the same revenue base.

What it means if you use Canva

For most users, nothing changes this month. The product keeps shipping, the free tier remains, and the pricing you pay today is unlikely to move because of a headcount decision. What does change, slowly, is the roadmap: fewer people on support, fewer on education and templates, and more of the engineering budget pointed at automated generation.

The practical risks for a small business are the ones that always follow a large reorganisation. Support response times get slower before they get faster. Documentation and template libraries go stale. Features that a specific team relied on can be deprecated quietly, because the team that understood why it existed is no longer there.

The sensible hedge is boring: keep your source files in a format you can move, do not build a workflow that only exists inside one vendor's automation, and export your brand assets to a folder you control. None of that is a comment on Canva's future. It is the same advice that applies to every tool you rent.

What to watch

Whether Canva publishes a breakdown by function in its next update, which would show whether engineering, support or marketing took the cut.
Whether the price of the paid tiers changes. A smaller company with the same AI infrastructure bill usually looks at pricing within two quarters.
Whether the free tier narrows. Free tiers are the first thing to be trimmed when a company needs to convert users rather than acquire them.
Whether other design platforms follow. If this is a structural read on the category rather than one company's correction, the next announcement will come within weeks.

For anyone working inside a design team, the useful question is not whether AI can do parts of your job. It is which parts of the job are judgement and which are production. The production half is being automated at every company at once. The judgement half is where the remaining roles are being concentrated, and the gap between the two is where this round of cuts landed.

Originally published at voxahq.in.

How to tell if text was written by AI (and what the labels actually mean)

VOXA — Tue, 03 Nov 2026 12:36:16 +0000

Most of the advice about spotting AI text is wrong, and it fails in both directions: it flags human writing and clears generated writing. Here is what actually holds up, written from the perspective of someone who publishes every day and has to check sources.

Why "it sounds like AI" is not a test

The signals people learn, like em dashes, tidy three-part lists, or the phrase "it is important to note", appear in plenty of human writing. They are artefacts of a style of editing, and style is copyable. Judging authorship from tone gives you a guess about how something was written, not who wrote it.

Detector tools are a different problem. They return a probability from patterns in text, and that score moves when text is edited, translated or lightly rewritten. A high score is not proof, a low score is not clearance.

The tells that survive contact with reality

Specifics that check out. Named sources, dates, version numbers, direct quotes with a link. AI can produce these, but it produces them wrong more often. Verifying three specific claims beats any stylistic read.
Claims that cannot be checked at all. "Studies show" with no study, "experts agree" with no expert. This is the most common pattern in low-effort generated copy.
Uniformity at length. Long text with no digressions, no corrections, no oddly specific personal detail, and identical paragraph lengths throughout.
Errors a human would catch. Citations to papers that do not exist, quotes attributed to the wrong person, products with features that never shipped. This is the strongest practical signal.
Provenance. C2PA content credentials, platform AI labels, and image metadata can confirm generation. Their absence proves nothing, because most tools strip them.

How to check a text in five minutes

Take the most checkable claim and verify it. If the piece cites a report, open the report. Two or three checks usually settle it, because fabrication clusters.
Look at the byline and date. A named author with a history, a publication date, and a corrections policy is a stronger signal than any stylistic marker.
Check whether the text admits uncertainty. Human writers hedge about what they do not know. Generated text tends to be confident everywhere, including about things that are wrong.

What AI labels do and do not mean

A label tells you a platform detected generation, or a creator disclosed it. Both are useful and both are incomplete. Platforms label unevenly, and creators can simply not disclose.

The practical rule: a label is evidence in one direction only. If it says AI, believe it. If there is no label, do not conclude a human wrote it.

Regulators are moving on this. The EU AI Act sets transparency duties for certain AI-generated content, and a Deloitte survey published this week found 84% of Australians want labelling to be a legal requirement, not a platform choice.

If you are the one publishing

Say what the machine did and what the person did. "Drafted with AI, checked and edited by a human" is a sentence readers can act on, and it costs nothing to write.

Keep sources visible, keep a named human accountable, and never let a tool invent a citation you did not open. That is the whole discipline, and it works better than any detector.

Originally published at voxahq.in. I write about AI tools and automation there without affiliate links in news or guides.

What AI voice costs per finished minute: three tools, three meters

VOXA — Mon, 02 Nov 2026 13:11:10 +0000

Three voice tools, three different meters. The sticker price tells you nothing until you convert it to a finished minute.

Tool	How it meters	Cost of one finished minute
ElevenLabs Creator	Credits, about 1,000 per minute	$22 covers about 121 minutes a month
Murf API	Characters, $0.01 per 1,000	Roughly $0.04 per minute
Speechify Premium	Flat subscription	$29 a month, unlimited listening

Prices checked on each vendor page, 2 November 2026. Character-to-minute conversion assumes about 1,000 characters per 15 seconds at normal pace.

What the gap means

If you generate ten minutes a month, ElevenLabs Creator costs $22 and Murf API costs about $0.40. The gap is not a discount, it is a different product: one sells a studio with cloning and quality tiers, the other sells characters.

So the question is not which is cheaper. It is whether you need the studio. For narration you will re-record, buy the subscription and stop counting. For generated clips at volume, count characters and pay per output.

The retake problem

A five-minute script is rarely recorded once. Every retake is charged again, and that is where plan maths breaks: budget roughly double your finished runtime. On ElevenLabs Creator that turns 121 minutes into about 60 minutes of publishable video.

Originally published at voxahq.in.

Best AI tools with a genuinely free plan in 2026

VOXA — Mon, 02 Nov 2026 13:10:41 +0000

A free plan is only useful if it lets you finish something. These six let you publish, ship or send without paying. Everything else in this category is a trial wearing the word free.

Tool	What free gets you	The catch
ChatGPT	Chat, file uploads, some image generation	Newest models sit behind $20
Gemini	Chat plus Google app integration	Limits on the strongest models
Claude	Long documents, careful writing	Tighter message limits on heavy days
Canva	Design editor, templates, 5GB storage	Most AI tools are Pro-only
Kit	Email to 10,000 subscribers	Automations need $33/mo
n8n	Unlimited workflows, self-hosted	You run the server

What free means in each case

Three business models hide behind the same word. Chatbots are free because usage costs are small and the habit is valuable. Newsletter platforms are free up to a subscriber count, because the vendor is buying your future list. Open-source tools are free because you supply the server, the updates and the backups.

Knowing which one you are using tells you when the bill arrives. A chatbot free tier rarely expires. A subscriber cap arrives with growth, which is a good problem. A self-hosted server arrives as a monthly invoice from your host, not from the vendor.

Where free plans quietly stop working

Voice, video and image generation are the worst offenders. They meter by output, so free tiers are measured in minutes or images rather than features, and one project can eat the month.

The second trap is commercial use. Several free tiers forbid it, which means the audio or image you generated cannot go into a monetised video. Read the licence before you build a workflow on top of a free plan.

What we did not test

We did not run a project end to end on all six, and free tiers change without notice. The descriptions reflect each vendor published plan on 2 November 2026.

Originally published at voxahq.in.

RWS closed the Acolad deal, and freelance translators lost another buyer

VOXA — Mon, 02 Nov 2026 13:10:06 +0000

RWS, a UK-listed language and AI company, completed its acquisition of Acogroup, the parent of Acolad, on 2 November 2026. Acolad is one of Europe's largest language services providers, with work in localisation, interpreting, transcription and medical-device documentation.

For the freelancers who supply that work, the practical effect is consolidation: one fewer large buyer, and more of the workflow moving into software the buyer controls.

What was announced

The release is short and corporate: the deal closed after "the conclusion of social consultation and all regulatory approvals". RWS says it strengthens its European and North American presence, and that Acolad's customers gain access to RWS's Cultural Intelligence Layer, Language Weaver Pro and Trados.

Acolad brings an established enterprise customer base and expertise in localisation, interpreting, transcription and data services, with strengths in regulated industries, public institutions and medical devices. RWS describes itself as serving 80+ of the world's top 100 brands, with 250,000+ data specialists and language experts behind its platform.

Why this matters outside the boardroom

Language services is one of the few industries where AI arrived with a measurable price per unit. Translation is billed by the word, interpreting by the hour, transcription by the minute. That makes it the cleanest place to watch what happens when a buyer can put a model and a human in the same workflow and decide, per job, which is cheaper.

Consolidation removes a layer of choice. A freelance translator who worked for both companies now works for one purchasing department, with one rate card and one vendor portal.

What the release does not say

There is no number of jobs affected, no plan for overlapping vendor lists, and no statement on rates. Those details usually arrive later, through consultation documents and contractor threads.

What to watch next

Whether Acolad's public-sector contracts are re-tendered or novated to RWS. Whether RWS publishes anything about its post-merge vendor programme. Whether freelance communities report rate changes, which usually shows up within a quarter.

For anyone working in language services, the useful move is boring: keep a second buyer, keep direct clients, and do not let one portal become the only route to work.

Originally published at voxahq.in.

ElevenLabs free plan: 10,000 credits is about ten minutes of audio

VOXA — Mon, 02 Nov 2026 12:38:27 +0000

Ten thousand credits sounds like a lot until you convert it: about ten minutes of finished speech a month.

What the free plan includes

Item	Free plan
Monthly credits	10,000
Roughly	10 minutes of generated speech
Projects in Studio	3
Commercial licence	No, that starts on Starter at $6

Checked on elevenlabs.io/pricing, 2 November 2026.

Why ten minutes runs out

A 1,500-word script is roughly ten minutes of speech, so one episode uses the whole month. Then there is editing: a line read wrong, a name pronounced badly, a pause in the wrong place. Every retake is charged again.

In practice, budget twice your finished runtime. A ten-minute video is twenty minutes of generation, and the free plan covers half of one video.

When $6 is the right answer

Starter raises the ceiling to 30,000 credits and adds the commercial licence. If the audio goes into a monetised video, a client project or a product, the licence is the reason to pay, not the credits. At $6 a month it is cheaper than the argument about whether you were allowed to publish it.

Originally published at voxahq.in.

What AI voice really costs per finished minute: ElevenLabs vs Murf vs Speechify

VOXA — Mon, 02 Nov 2026 12:37:58 +0000

Three voice tools, three different meters. The sticker price tells you nothing until you convert it to a finished minute of audio.

I checked each vendor's pricing page on 2 November 2026. Here is what they actually cost.

The comparison

Tool	Price	Best at	Where it hurts
ElevenLabs	Free $0, Starter $6, Creator $22, Pro $99	Natural narration for video	Credits burn fast when you re-record
Murf	API from $0.01 per 1,000 characters	Volume and automation	You think in characters, not minutes
Speechify	Free, Premium $29/mo or $139/yr	Turning documents into audio	Built for listening, not cinematic narration

The only maths that matters

Voice tools meter differently, which makes the headline price useless. ElevenLabs sells credits, roughly 1,000 per minute of speech. Murf sells characters, and 1,000 characters is about 15 to 18 seconds at a normal pace. Speechify sells a subscription for listening.

Compare at the point your file is finished. A five-minute video needs about 5,000 ElevenLabs credits, or roughly 18,000 characters on Murf, which at $0.01 per 1,000 characters is about $0.18 of API usage. That gap is the whole story.

ElevenLabs: you pay for the plan, not the minutes

Starter at $6 covers 30,000 credits and the commercial licence. Creator at $22 raises that to 121,000 and adds professional voice cloning. Pro at $99 lifts output quality and gives 600,000 credits.

The trap is editing. A five-minute script is rarely recorded once, and every retake comes out of the same credit bucket. Budget roughly double your finished runtime.

Murf: the per-character option

Murf's API pricing is the most transparent here: $0.01 per 1,000 characters for text to speech, $0.03 for speech to text, $0.10 per minute for translation. You are buying output, not a seat.

That suits a pipeline generating many short clips, and a business that wants to forecast spend per project. It suits creative narration less well, because you lose the flat monthly ceiling.

Speechify: a reading tool that also generates audio

Free gives you ten voices and playback up to 1.5x. Premium at $29 a month, or $139 billed yearly, unlocks 1,000+ voices, 60+ languages, 5x playback, voice typing and AI podcasts. Excellent for turning a PDF into something you can listen to; not the tool for a 40-second commercial read.

What we did not test

We did not generate the same script on all three and compare the audio, and we have not checked licences case by case. Every price comes from the vendor's pricing page on 2 November 2026. Generate one paragraph on each free tier and listen before you pay.

Originally published at voxahq.in.

Best AI tools for writing email newsletters: what each platform really costs

VOXA — Sun, 01 Nov 2026 14:15:35 +0000

beehiiv and Kit are the two cheapest ways to run a newsletter with AI help, and both have a free tier that is real. Ghost is the pick when you want the newsletter and the website in one place and do not mind paying $18 a month.

Prices below come from the vendors' own pricing pages on 1 November 2026.

Which newsletter tool should a small team pick?

Tool	Price	Good at	Where it hurts
Kit	Free up to 10,000 subscribers; paid from $33/mo yearly	A free tier that stays usable for years	Plainer landing pages
beehiiv	Free plan with unlimited sends; Lite $49/mo monthly, $39/mo yearly	Growth: referrals, recommendations, ad network	Paid plans cost more than Kit
Ghost (hosted)	$18/mo billed yearly (Starter); $29/mo Publisher	Newsletter plus a real website on one domain	No built-in growth tools
Ghost (self-hosted)	Software free; you pay hosting	Lowest long-run cost if you can run a server	You own updates, backups, deliverability

What "AI" actually does in a newsletter tool

Almost every platform advertises AI. In practice it writes subject line variants, drafts an email from your notes, and summarises what performed last month. None of that is worth paying double for.

What is worth paying for is the list: deliverability, a clean editor, and a form that converts. Choose on those, then use a $20 chat subscription for the writing if the platform's AI is weak.

Kit: free up to 10,000 subscribers

Kit's free plan covers 10,000 subscribers, which is more than most newsletters will ever need. You can run for two years and pay nothing. Paid plans start at $33 a month billed yearly ($390); you pay for automations, sequences and a referral program, not for basic sending.

beehiiv: free sending, paid growth

beehiiv's free plan includes unlimited sends, templates, landing pages and an API. The paid Lite plan is $49 a month, or $39 billed yearly, and adds paid subscriptions, ad monetisation and automations. It is the better pick if you intend to grow with referrals, and the worse pick if you just want to send a weekly email cheaply.

Ghost: newsletter and website in one

Ghost Starter is $18 a month billed yearly and gives you a newsletter plus a real website on your own domain. Publisher is $29, Business $199. If you already pay for a site and a newsletter separately, Ghost replaces both. Self-hosting the same software is free and you pay only for a server, but updates, backups and deliverability become yours.

The setup that costs nothing for a year

Start on Kit's free plan, write in whatever AI tool you already pay for, and move only when you hit a paid feature you need. The mistake is paying $39 to $49 a month before the list has 100 engaged readers. The platform is not what makes a newsletter grow; the writing and the consistency do.

What we did not test

We did not run a live campaign on all four platforms and are not claiming inbox placement numbers. Every price comes from the vendor's pricing page on 1 November 2026.

Originally published at voxahq.in.

Best AI tools for writing product descriptions: what each one really costs

VOXA — Sun, 01 Nov 2026 14:15:02 +0000

Most stores do not need a $59 a month AI tool to write product descriptions. They need a spec sheet and a $20 assistant. The rest is a pricing decision that depends on one question: how many people write for your brand?

I checked the pricing pages for six tools on 1 November 2026. Here is what each costs, and where it stops being worth it.

The prices, side by side

Tool	Price	Good at	Where it hurts
ChatGPT Plus	$20/mo	Bulk descriptions when you paste your own specs	No catalog awareness
Claude Pro	$20/mo, about $17 yearly	Following a long style rule sheet	Usage caps on long runs
Rytr Lite	$9/mo	Volume at the lowest price	Generic output, rewriting is on you
Copy.ai	$29/mo ($24 yearly)	Cheap chat across several models	Real automation starts in four figures
Jasper Pro	$59 per seat/mo	Keeping several writers on one voice	Expensive for one person
Shopify Magic	Included in your Shopify plan	First drafts inside the product editor	Shopify only, no bulk export

That spread, from $9 to $59, is not a quality spread. It is a difference in who the tool is built for.

What a product description actually has to do

A product page answers three questions before someone adds to cart: what is this, will it work for my case, and what will annoy me later. A general model gets the first one right and guesses at the other two, which is why raw output reads like every other store.

The fix is not a better model. It is feeding the model the spec sheet: dimensions, materials, what is in the box, who it is not for. Every tool below improves the moment you paste that in, and no tool fixes a missing spec sheet.

ChatGPT Plus at $20: the default

For a store under 200 products, start here. Paste the specs, ask for three lengths, pick one. Its weak point is memory across a catalog: you will re-paste the brand rules every session unless you keep a saved prompt. That is a workflow problem and it is free to fix.

Claude Pro at $20: better at rules

Claude holds a long rule sheet better: banned words, sentence length, how to describe materials, claims to avoid. If descriptions have to pass a brand or legal review, this is the one that keeps the rules in place for a whole run. The trade-off is usage limits on heavy days.

Rytr from $9: the cheapest volume

Lite is $9, Personal $19, Pro $49. For 50 simple products and no brand voice to protect, $9 is enough. What you give up is judgement: clean sentences, little specificity, so every description needs a human pass.

Copy.ai at $29: a chat tool wearing an automation label

Chat across several models for $29 sounds like a bargain until you want automation. The workflow product is priced for teams and starts in four figures a month. Treat it as a $29 chat subscription.

Jasper Pro at $59 per seat: for teams

Jasper's pitch is brand voice shared across seats. That is useful when five people write listings, and pointless when one person does. The writing is not six times better than a $20 assistant.

Shopify Magic: already paid for

If you sell on Shopify this is included in your plan. It writes a first draft inside the product editor. Plain but usable, zero extra cost, and it will not learn your voice from previous products.

The setup that costs $20 and works

One assistant plan plus a saved prompt covers most stores. Write the prompt once: audience, banned claims, sentence length, the three questions above, and one example you are happy with. Then paste each spec sheet and ask for two variants.

Move up to Jasper only when a second writer joins, and down to Rytr only when the catalog is simple enough that generic copy is acceptable.

What this comparison does not cover

I did not run a live store and did not measure conversion rates. Every price comes from the vendor's pricing page on 1 November 2026. Prices change, so check the vendor page before buying.

Originally published at voxahq.in.

Rytr at $9 vs Jasper at $59: which one a small store needs

VOXA — Sun, 01 Nov 2026 14:14:02 +0000

Rytr Lite is $9 a month. Jasper Pro is $59 a seat. Both write product copy, and the gap is not six times better writing. It is who else has to use it.

The prices, plainly

Tool	Price	What that buys
Rytr Lite	$9/mo	Unlimited characters, one person, generic output
Rytr Personal	$19/mo	More sources and research runs
Rytr Pro	$49/mo	Higher limits and connections
Jasper Pro	$59 per seat/mo, billed yearly	Brand voice shared across seats

Rytr's numbers are from rytr.me/pricing, Jasper's from jasper.ai/pricing, both checked on 1 November 2026.

What the $50 difference actually is

Rytr is a generator: you give it a product, it gives you sentences. At $9 that is a fair trade, and for a store with simple products it is all you need.

Jasper is a workspace with a voice layer. You define the brand once, and every seat produces copy that sounds like the same company. That matters when five people write listings, and it is worthless when one person does.

The real cost nobody prices

Cheap generation is not free if every description needs a rewrite. At $9 a month with 300 products, you will spend more hours editing than the subscription saves. At $59 with one writer, you are paying for a feature you cannot use.

The honest split: one writer, under 200 products, Rytr or a $20 chat plan. Two or more writers sharing a brand, Jasper's per-seat price starts to make sense.

This note first appeared on voxahq.in, where the prices are checked by hand.