DEV Community: Soluma AI

AI Automation Cost: Why $0.12 Per Run Can Become $12,000 at Scale

Soluma AI — Fri, 06 Nov 2026 07:13:10 +0000

Originally published on Soluma AI.

One of our workflows cost about $0.12 per prospect across a 500 run test. That is either trivial or a $12,000 decision, depending entirely on the multiplier behind it.

One of the most useful lessons we learned about AI cost started with a number that looked almost too small to care about: $0.12. That was roughly what one of our AI workflows was costing to research a prospect. Twelve cents. For a workflow doing real research, that sounded perfectly reasonable, and at small volume it was.

Then we did the obvious calculation we probably should all do much earlier when building AI systems. What happens at 100,000 runs? The answer is $12,000. Suddenly, a fraction of a dollar is no longer a technical detail. It is an operating decision.

That led us to look much more carefully at what the $0.12 actually contained. The workflow was not making one simple AI request. It was carrying out six separate buyer searches, processing the findings and performing a final analysis. The interesting question was never which AI model is cheapest. It was what one completed piece of work actually costs, where that money goes, and how much can be removed before the work gets worse.

How much does AI automation actually cost per run?

There is no universally good AI automation cost per run. The useful number is the cost of producing one acceptable completed job, including model usage, tool calls, search or retrieval, external APIs, retries and other resources required by the workflow. A $0.12 workflow may be economical at 500 runs and material at 100,000. The cost only makes sense when it is viewed alongside volume, quality and the commercial value of the output.

Quick takeaways

1Tiny unit costs become real budgets at volume: $0.12 per run becomes $12,000 across 100,000 runs.
2Model price is only part of the cost. Search, retrieval, tools, retries and workflow design create a cost floor that changing models does not remove.
3Cheaper output is not automatically better economics. If saving a few cents reduces research quality, the cheaper workflow may create less business value.
4Measure before you optimise. Establish a quality baseline, calculate the full cost per acceptable job, then change one part of the workflow at a time.

The $0.12 problem: tiny number, real money

Here is what the same workflow looks like at different volumes. Nothing changed technically between the first row and the last. Only the multiplier changed.

RUNS	COST AT $0.12 PER RUN
100	$12
500	$60
1,000	$120
10,000	$1,200
50,000	$6,000
100,000	$12,000

This is one of the easiest things to miss while prototyping. During development you might test ten rows, twenty, fifty, so a workflow can feel almost free. Then someone adds a Run All button, and that button changes the economics completely. If 100,000 eligible records sit behind it, a twelve cent workflow has turned one click into roughly $12,000 of authorised usage.

The question is not whether $0.12 is cheap. It is how many times you are about to spend $0.12.

Model cost and workflow cost are not the same thing

A common way to estimate AI cost is to read the model pricing page: input tokens cost X, output tokens cost Y, multiply and you have your number. That works for a single model request. It becomes incomplete once the model sits inside a larger workflow.

web searches and retrieval
external APIs and database operations
multiple model calls and reasoning steps
validation calls and secondary models
retries and failed attempts

OpenAI, for example, currently prices its web search tool at $10 per 1,000 calls, or $0.01 per search call, with search content tokens additionally billed at the selected model rates. That mattered in our workflow. Each prospect required six buyer searches, so six search calls alone contributed six cents before any analysis. That is half of a $0.12 run in tool call fees.

This is what we mean by a structural cost floor. If the workflow genuinely requires those six searches, switching to a cheaper analysis model does not make that six cent layer disappear.

Our 500 run test changed what we wanted to optimise

Before scaling further we ran the workflow across approximately 500 prospects, roughly $60 in operating cost. We were not testing cost alone. We wanted to know whether the searches produced useful evidence, whether the buyer questions were realistic, whether the final analysis retained the right context, whether outputs were consistent enough to use downstream, and where failures happened.

If we had optimised for price before establishing a quality baseline, we would have had no reliable way to tell whether a cheaper version was still doing the same job. We did test cheaper configurations. The cost went down and so did the quality. That told us some of the spend was not waste. It was buying information the workflow needed.

Cost per run can still hide the real number

Even after calculating cost per run, there is another level worth measuring: cost per acceptable completed job. Suppose a workflow costs $0.12 every time it starts. If only 90 percent of outputs meet the acceptance standard, you did not pay twelve cents for every usable result. You paid more.

Cost per acceptable completed job = total workflow cost divided by accepted outputs.

ACCEPTANCE RATE	EFFECTIVE COST PER ACCEPTED RESULT
100%	$0.120
95%	$0.126
90%	$0.133
80%	$0.150

This is why an apparently cheaper configuration can lose its advantage quickly if it creates more failures, retries or unusable outputs. The business is not buying API calls. It is buying completed work.

The cheapest model does not necessarily create the cheapest system

Model choice obviously matters, and current model line ups reflect that trade off between frontier capability, balanced cost and high volume workloads. But a cheaper model only saves money on the work that model is responsible for. It does not reduce search calls, external API fees, duplicate retrieval, unnecessary workflow stages, poor retry logic, or calls that should never have happened.

OPTIMISATION LAYER	WHAT YOU CHANGE	TYPICAL EXAMPLES
Model optimisation	How expensive each model operation is	Smaller model, lower reasoning effort, shorter output, prompt caching
Architecture optimisation	How much work the system performs	Fewer calls, reused results, conditional research, deterministic processing, controlled retries

Both matter, but the second becomes increasingly important as volume rises. If your system performs eight paid actions where four would produce the same result, changing models treats the symptom while the workflow keeps doing unnecessary work.

The better question: what is the lowest cost that preserves the job?

There is a dangerous version of AI optimisation that looks excellent on a spreadsheet. Version A costs $0.12. Version B costs $0.08. Success. Except Version B misses more evidence, produces weaker classifications, requires more human checking or sends poorer output into the next stage. The four cent saving is real. So is the quality loss.

For prospect research that influences who a business contacts and what it says to them, four cents is an absurd place to compromise when the downstream opportunity is worth hundreds or thousands of dollars. A better target is the lowest cost at which the workflow still performs its job to the required standard.

What we are optimising next

Our approximately $0.12 figure is the baseline, not the finish line. The working target we are testing towards is around $0.08 per run while maintaining the same output quality. That is not a market benchmark. Another workflow could reasonably cost $0.01, $0.40 or $4 depending on what it does.

VOLUME	$0.12 PER RUN	$0.08 PER RUN	POTENTIAL SAVING
500	$60	$40	$20
10,000	$1,200	$800	$400
50,000	$6,000	$4,000	$2,000
100,000	$12,000	$8,000	$4,000

At 500 runs the difference is $20 and easy to dismiss. At 100,000 runs the exact same engineering improvement is worth $4,000. That is why we test small but design with the larger volume in mind.

Where should you look for AI cost savings first?

1. Establish the quality baseline

Take a representative sample and run the version you currently trust. Define what a good output means before changing anything. Accuracy alone may not be enough. Depending on the task you may also care about evidence quality, completeness, consistency, format compliance, latency or human review time.

2. Break the run into cost components

Do not leave yourself with one number called AI cost. Separate model input, model output, tool calls, retrieval and search, external APIs, retries, failures and infrastructure. Now you know what is actually worth optimising.

3. Find the structural floor

Ask which costs remain even if you changed the model tomorrow. In our case, six paid buyer searches immediately exposed a meaningful part of the floor, which changes where engineering effort should go.

4. Remove duplicated work before reducing intelligence

Could a result already collected in step two be reused in step five? Is static context being repeatedly sent? Could deterministic code handle filtering, joining, ranking, deduplication or validation instead of another model call? Current OpenAI guidance recommends bounded programmatic processing for predictable operations rather than forcing every intermediate step through a model.

5. Use cheaper models where the task genuinely allows it

Classification does not need the same model as final synthesis, and extraction does not need the same reasoning level as strategic analysis. Current guidance recommends evaluating model tiers and reasoning levels against representative tasks. The key word is evaluating. Do not downgrade and assume equivalence. Test it.

6. Make prompts leaner, not merely shorter

A shorter prompt only helps if it still contains the instructions that affect the result. Remove repeated instructions, irrelevant tools and unnecessary examples, while retaining guidance that corrects measured quality gaps. The goal is not prompt minimalism. It is removing instructions that are not earning their cost.

7. Use caching where repeated context makes it worthwhile

If large parts of the prompt stay identical across runs, prompt caching can reduce the cost of reprocessing the same input. Measure actual cache hits. A feature being available does not mean your workflow is structured to benefit from it.

8. Consider batch processing where the workload fits

For asynchronous work that does not need an immediate response, the OpenAI Batch API currently offers a 50 percent discount within a 24 hour completion window. That can be meaningful at volume, though tool heavy workflows should be tested for compatibility rather than moved wholesale.

9. Control retries and stopping conditions

Retries are useful when something genuinely failed. They are expensive when nobody told the system when to stop. Define what counts as success, recoverable failure, permanent failure and stop. This matters even more when the workflow can run without anyone watching.

What the Run All button really means

A button labelled Run All Empty Rows looks operational. Financially it is closer to: authorise every eligible row to consume the resources required to finish. Two hundred rows at $0.12 is about $24. Ten thousand is about $1,200. One hundred thousand is about $12,000.

This is not an argument against automation. It is an argument for attaching economics to automation. Before enabling large autonomous queues, a production system should know how many records are eligible, the expected cost per record, the likely failure rate, the maximum acceptable spend and how execution can be stopped.

Four numbers we now want before scaling an AI workflow

1Cost per acceptable completed job, not just the model request or the tokens.
2Structural cost floor, meaning the costs inherent in the current architecture.
3Expected volume, because a unit cost without volume is almost meaningless.
4Value of the output, because cost must be judged against the value created.

A four cent saving on an internal classification task could matter enormously across millions of runs. A four cent saving on high value B2B research may matter far less than preserving the quality of the research.

What people often misunderstand about AI cost optimisation

The biggest misunderstanding is treating optimisation as a race towards the cheapest possible model. Sometimes the model is the problem and sometimes it is not. The useful question is which part of this workflow is expensive without improving the final result. That could be a model, a redundant search, duplicated context, excessive reasoning, a needless intermediate summary, poor retry logic or a human review step caused by unreliable output upstream.

Removing useful intelligence is cost cutting. Removing unnecessary work is optimisation. They are not the same thing.

When cost optimisation may not be worth doing yet

If a workflow runs 200 times per year and costs a few dollars in total, spending days redesigning it to save two cents per run is poor economics. Likewise, if a workflow is still changing dramatically, premature cost engineering may optimise architecture that disappears next week. Prove the job works, measure it, understand the cost, then optimise what survives.

High volume repeatable workflows are different. At 100,000 runs, a one cent reduction is worth $1,000 and a four cent reduction is worth $4,000. Now the engineering conversation changes.

Questions to ask before scaling an AI workflow

What does one acceptable completed job cost?
Which part of that cost comes from models, and which from tools, search, APIs or retrieval?
What is the unavoidable cost floor?
What percentage of runs produce acceptable outputs, and what happens when something fails?
Can the workflow retry indefinitely?
Is repeated information being fetched or processed more than once?
Could simpler steps use cheaper models or deterministic code?
How much quality can change before the optimisation stops being worthwhile?
What is the expected monthly or annual volume, and what is a successful output worth commercially?

If those answers are unknown, Run All is still partly a guess.

How should businesses think about value, not just API price?

An AI workflow is economically useful when the value of the work it produces comfortably exceeds the cost of producing and operating it. Model pricing makes it easy to become obsessed with decimal places. If high quality research costs twelve cents and materially improves a high value sales process, twelve cents may be excellent economics. If a workflow performs a low value task ten million times, twelve cents may be completely unacceptable.

The lesson from twelve cents

The important thing we learned was not that our workflow costs $0.12. That number will change. Models change, pricing changes and the architecture will change. The more useful lesson is how we now think about scaling: we test before we multiply, we establish what good output looks like, we measure where the money is actually going, then we optimise the system without quietly removing the things that made it useful.

An AI demo only has to work. A production AI system has to work repeatedly, economically and at the quality the business actually needs. That is how we approach the systems we build with clients at Soluma: start with the business problem, test the workflow under real conditions, measure the economics, then improve the architecture before volume magnifies every unnecessary decision.

Frequently asked questions

What is a good cost per run for AI automation?

There is no universal good cost per run. A useful target depends on task complexity, volume, success rate and the value of the finished output. The correct comparison is between cost and business value, not between your workflow and an arbitrary industry average.

How do I calculate the real cost of an AI workflow?

Add model token costs, tool and search fees, external APIs, retrieval, retries, failed runs and other operating resources, then divide total spend by the number of acceptable completed outputs. This gives a more useful figure than model cost alone.

Are web search calls charged separately from model tokens?

Yes, for OpenAI's current web search tool. OpenAI lists web search at $10 per 1,000 calls, with search content tokens additionally billed at the selected model rates.

Is using a cheaper AI model always the best way to reduce cost?

No. A cheaper model helps only if it maintains the required quality and reliability. Workflow architecture, duplicate calls, unnecessary retrieval, retries and excessive context can sometimes offer better savings without weakening the result.

When should I start optimising AI costs?

Start measuring costs during testing, but make major optimisation decisions after you have a reliable quality baseline. Otherwise you risk reducing cost without knowing whether you also reduced the value of the output.

How do I stop an AI automation from creating a surprise bill?

Set clear run limits, budget thresholds, stopping conditions, retry limits and cost monitoring before allowing large autonomous queues. Forecast expected spend by multiplying eligible volume by measured cost per acceptable completed task.

What is the difference between cost per run and cost per completed task?

Cost per run measures one execution attempt. Cost per completed task measures what you actually spent to obtain a usable result after accounting for failures, rejected outputs and retries. For business decisions, the second number is usually more useful.

This article first appeared on soluma.ai.

Talk to Soluma AI

Soluma AI · AI agency and consulting firm · Singapore, Malaysia, Australia, New Zealand, United States, Canada, United Kingdom, UAE

AI Search Engine Optimization: How to Get Your Brand Found, Cited and Recommended in AI Search

Soluma AI — Fri, 06 Nov 2026 05:11:43 +0000

Originally published on Soluma AI.

Written by: Soluma Editorial Team Article type: Research-backed practical guide Last updated: October 8, 2026

Your website can rank on Google and still disappear from an AI-generated answer.

That is the uncomfortable part of AI search.

A buyer asks ChatGPT, Google AI Mode, Copilot, Gemini or another AI system which provider to consider. The answer arrives in seconds. Three competitors are named. Two industry publications are cited. One comparison page influences the recommendation.

Your company is nowhere in it.

The obvious reaction is to ask:

“How do we optimize our content for AI?”

But content formatting is only part of the problem.

Before an AI system can cite or recommend a company, several things have to go right. Relevant information needs to be discoverable. The business has to be understandable. Claims need enough evidence to be credible. Useful passages have to be retrievable. Information elsewhere on the web may influence how the brand is interpreted. And after all of that, the system still has to decide that your company belongs in the answer.

That is what AI search engine optimization needs to address.

Not another checklist of tricks.

A system for becoming findable, citable and commercially relevant when buyers ask AI what to do, what to buy and who to choose.

What is AI search engine optimization?

AI search engine optimization is the process of improving how easily AI-powered search and answer systems can discover, understand, retrieve, cite and accurately represent a brand and its information. It includes traditional SEO foundations, but adds work around conversational buyer questions, evidence, entity clarity, third-party corroboration and AI-specific measurement.

The objective is not merely to earn a citation. For commercial businesses, the stronger outcome is being correctly included when a buyer asks which solution, product or provider should be considered.

Quick Takeaways

SEO is still the foundation: AI search cannot consistently retrieve information that is difficult to crawl, index or understand.
Citations and recommendations are different outcomes: Your content can help an AI answer while your competitors still receive the commercial recommendation.
Optimize around buyer decisions, not just keywords: AI conversations combine needs, constraints and comparisons in ways that traditional keyword databases often fail to capture fully.
Measure before producing more content: Establish where your brand appears, which competitors win and which sources influence the answers before deciding what to change.

AI search optimization is bigger than “getting cited”

The phrase AI search visibility tends to collapse several different outcomes into one number.

That can be misleading.

Imagine a potential customer asks:

“Which CRM should a 30-person B2B company shortlist if it needs strong automation but does not have a dedicated CRM administrator?”

An AI answer recommends:

Company A Company B Company C

Then cites an article from your website because your article contains a useful CRM implementation statistic.

Did your GEO strategy succeed?

Technically, you earned a citation.

Commercially, perhaps not.

Your information contributed to the answer.

Your company did not enter the shortlist.

This is why we separate three outcomes.

Visibility type	What happened	Commercial meaning
Source visibility	Your website or research is cited	The system found your information useful
Brand visibility	Your company is explicitly named	The system associates your brand with the topic
Consideration visibility	Your company is recommended, shortlisted or compared as an option	You entered the buyer's decision set

All three matter.

But they should not be treated as equivalent.

For a publisher, source visibility may be the primary objective.

For a B2B service company, software provider, clinic, professional-services firm or retailer, consideration visibility may be worth considerably more.

This is one reason AI search optimization should not begin with rewriting paragraphs.

It should begin with understanding which outcome the business actually needs.

How do AI search systems find information?

A single buyer question can trigger several retrieval and comparison steps before an AI system constructs its answer.

Different AI platforms use different models, indexes, search providers, retrieval systems and citation processes, so there is no single universal architecture.

But for retrieval-based AI search, a simplified flow often looks like this:

User question → question interpretation → related searches or retrieval → candidate sources → relevant passages or facts → synthesis → citations / mentions / recommendations

Google gives us one useful window into this process.

Its generative Search systems use retrieval-augmented generation (RAG) connected to Google's Search index. Google also uses query fan-out, where a complex request can trigger several related searches to gather the information needed to answer it.

For example, a user may ask:

“What is the best accounting software for a Singapore SME with five employees that invoices internationally and wants low monthly costs?”

One question can imply several information needs:

accounting software for small companies
international invoicing support
Singapore tax compatibility
pricing
ease of use
implementation effort
alternatives

That is why optimizing only for one keyword can miss the real decision being made.

OpenAI's publisher guidance also confirms the technical discovery layer for ChatGPT Search: public sites can appear, and allowing OAI-SearchBot helps make content eligible to be discovered, surfaced and cited in Search results.

The exact mechanics differ by platform.

The strategic lesson does not:

AI visibility begins before the generated answer. It begins with discovery and retrieval.

The Soluma AI Search Visibility model

Different AI visibility problems require different interventions; diagnosis should come before content production.

We find it more useful to think about AI Search Optimization as six sequential questions.

1. Found

Can the relevant system access and retrieve information about you?

This is the technical layer.

2. Understood

Can it clearly determine what your company does, who it serves and what your offering is relevant for?

This is the entity and information architecture layer.

3. Trusted

Is there credible evidence behind your claims?

This is the evidence and authority layer.

4. Cited

Does your information contain something useful enough to support the answer?

This is the information-value layer.

5. Considered

When the question becomes “who should I choose?”, does your brand enter the shortlist?

This is the commercial visibility layer.

6. Measured

Do you know which of the previous five are actually improving?

This is the performance layer.

The sequence matters.

A company that cannot be found should not begin by obsessing over FAQ wording.

A company that is cited but never recommended has a different problem from one that is technically invisible.

Diagnosis determines the work.

Step 1: Make sure AI search systems can actually find the content

Before changing your writing style, check accessibility.

For Google, the company says pages must be indexed and eligible to appear with snippets to participate in its generative Search features. Existing crawling, indexing, canonicalization and JavaScript SEO practices remain relevant.

For ChatGPT Search, OpenAI says publishers wanting content eligible for summaries and snippets should not block OAI-SearchBot. It distinguishes OAI-SearchBot from GPTBot, which is related to potential model training.

For Microsoft surfaces, Bing Webmaster Tools and normal indexability remain relevant, and Microsoft recommends maintaining fresh, accessible pages and using IndexNow where appropriate to notify participating engines about updates.

A technical AI visibility audit should therefore examine practical questions such as:

Are important pages indexed?
Are important crawlers blocked?
Are canonical signals consistent?
Is crucial content accessible in HTML?
Does JavaScript prevent key content from being reliably processed?
Are duplicate URLs creating conflicting versions?
Are important updates being discovered quickly?
Are product, location and service facts consistent?

This is not glamorous.

It is also the layer every “AI citation hack” depends on.

Step 2: Stop treating keyword research as the whole buyer-question universe

Keyword research still matters.

But AI prompting creates a different form of demand data.

Semrush analyzed ChatGPT clickstream and prompt data and compared those prompts against a database containing more than 27 billion traditional search keywords. During much of the study period, 65% to 85% of the prompts could not be matched to a traditional search keyword in its database.

That does not mean keyword research has become useless.

It means a keyword database may not fully represent how a buyer speaks when given unlimited room to explain a problem.

Compare:

Google-style search

best CRM for B2B

with:

AI-style research

“We have a 25-person sales team across Singapore and Australia. We need automated follow-up, easy HubSpot migration and strong reporting, but we don't want a system that requires a full-time administrator. Which CRMs should we shortlist and why?”

The second question contains:

company size,

geography,

existing technology,

operational constraints,

desired features,

implementation concerns,

and an explicit request for recommendation.

That is not simply a longer keyword.

It is a decision scenario.

Build a buyer-question map

Instead of starting with 500 content keywords, map the questions around the buyer journey:

Problem recognition

Why are our leads not converting?

Solution exploration

What systems can automate lead follow-up?

Category education

CRM automation vs sales engagement platform

Evaluation

Best CRM for a 30-person B2B company

Comparison

HubSpot vs Salesforce for a growing regional business

Risk

What usually goes wrong during CRM implementation?

Cost

What does CRM implementation cost for a mid-sized business?

Selection

Which provider should we contact?

The objective is not to publish one page for every conceivable prompt.

Google explicitly warns against producing large numbers of pages purely to capture every fan-out or long-tail variation.

The objective is to understand the decision territory.

Then build genuinely useful coverage around it.

Step 3: Create information AI systems have a reason to retrieve

Generic information has become extremely cheap.

That changes the value of content.

An article titled:

“10 Benefits of AI Automation”

that repeats the same ten points found on hundreds of other websites creates very little new information.

Google's 2026 guidance specifically recommends non-commodity content: material grounded in firsthand experience, expertise, original analysis or a distinct point of view rather than another summary of information already available elsewhere.

This is a major AI Search Optimization opportunity.

Useful assets include:

Original datasets
Benchmark studies
Pricing analysis
Experiments
Implementation lessons
Methodologies
Calculators
Decision frameworks
First-party operating data
Detailed case evidence
Original expert interviews
Market-specific analysis
Transparent comparisons
Product specifications
Failure analysis
Firsthand reviews
Proprietary models

Consider the difference.

Generic claim:

AI automation can reduce operational costs.

Higher-information claim:

We tested a prospect-research workflow across 500 runs at approximately $0.12 per completed run, then identified where structural search costs prevented simple model switching from solving the economics.

The second contains information someone else cannot recreate merely by asking an AI to summarize the internet.

That makes it more valuable to the reader.

Potentially, it also makes it more valuable as a source.

Step 4: Make important information easy to extract without writing like a machine

There is good advice and bad advice around “AI-friendly formatting.”

Good advice says:

Make important information clear.

Bad advice says:

Rewrite the entire website into two-sentence chunks because LLMs cannot understand normal prose.

Google explicitly rejects the idea that pages must be artificially “chunked” for its generative Search systems. It also says websites do not need to rewrite everything using special AI phrasing.

Microsoft's guidance is more tactical: it recommends clear headings, concise self-contained sections, useful Q&A formats, lists and tables when those structures make information easier to interpret and reuse accurately.

Those positions are not contradictory.

The sensible middle ground is:

Write naturally, but make important information unambiguous.

For example:

Weak

Our advanced approach gives businesses the confidence to grow using innovative, future-ready capabilities.

What does that actually mean?

Stronger

Soluma's AI Search Visibility work identifies the buyer questions that matter, measures where a brand appears across AI-generated answers, maps competing brands and cited sources, then prioritizes the technical, content and authority gaps behind those results.

Now the entity, service and function are clearer.

Use the format that fits the information

Use a paragraph for nuance.

Use a table for comparison.

Use bullets for genuine lists.

Use numbered steps for a process.

Use a direct answer when the reader asks a direct question.

Use long-form explanation when the decision genuinely requires depth.

The goal is not to make content robotic.

It is to make meaning difficult to misinterpret.

Step 5: Give your claims evidence

AI Search Optimization becomes weak very quickly when every page makes claims without evidence.

Best-in-class.

Industry-leading.

Innovative.

Trusted.

Advanced.

These words tell a retrieval system—and a buyer—almost nothing.

Evidence is more useful.

Microsoft's AI Search guidance explicitly recommends supporting claims with examples, data and cited sources because evidence improves clarity and trust when material is reused in AI-generated answers.

The original academic work that formalized Generative Engine Optimization, presented at KDD 2024, tested several content interventions and reported visibility improvements of up to 40% within its experimental benchmark. It also found that effectiveness varied by domain.

That finding is frequently exaggerated online.

It does not mean:

“Add citations and your ChatGPT visibility increases 40%.”

The study measured visibility within a controlled generative-engine benchmark. It did not guarantee live-platform rankings, traffic or conversions.

The useful principle is simpler:

Claims become stronger when the reader can verify them.

Useful evidence can include:

Government data
Official documentation
Peer-reviewed research
Industry reports
First-party data
Clearly explained calculations
Named methodologies
Genuine customer evidence
Product specifications
Relevant expert commentary

And when evidence is uncertain, say so.

Trust often increases when an article explains what is not yet known.

Step 6: Make the business itself easier to understand

AI search optimization is not only about informational articles.

A buyer may ask:

“Which companies provide AI Search Visibility services in APAC?”

Now the system is not merely trying to retrieve a fact.

It is trying to understand business entities.

What does each company do?

What market does it serve?

What are its services?

Who does it help?

What evidence supports that positioning?

Where else is the company described?

That makes commercial pages more important than many GEO strategies acknowledge.

Check whether your website clearly explains:

Who you are

What you do

Who you help

What problems you solve

What specific services or products you provide

What differentiates the offering

Where you operate

What evidence supports your claims

A homepage full of language such as:

“Reimagining the future through intelligent innovation”

creates far more ambiguity than:

“Soluma helps businesses improve how they are found, cited and considered in AI-powered search.”

The second is less poetic.

It is considerably easier to understand.

Step 7: Build authority beyond your own website

A company cannot define its entire reputation by publishing claims about itself.

AI-generated answers may use information from:

industry publications,

news sites,

review platforms,

directories,

professional communities,

videos,

forums,

partners,

analyst content,

customer reviews,

and other third-party sources.

Google confirms that its generative Search experiences can surface information about products and services from sources across the web, including blogs, videos and forums. But it also explicitly warns against pursuing inauthentic mentions as an optimization tactic.

That distinction is critical.

Good off-site authority

A company conducts useful research that earns media coverage.

A founder contributes expertise to an industry publication.

Real customers leave genuine reviews.

A specialist publication independently compares the service.

Industry partners reference the company.

Relevant communities discuss the company because people genuinely use it.

Bad off-site authority

Fake Reddit conversations.

Paid “best companies” pages that exist only for links.

Manufactured Quora answers.

Fake user accounts recommending the brand.

Mass-produced PR pages containing identical claims.

The objective is not:

“Get our name on as many websites as possible.”

It is:

“Create enough credible external evidence that buyers and retrieval systems can independently understand where our company belongs.”

That takes longer.

It is also considerably harder for competitors to copy.

Step 8: Keep important facts consistent and current

AI search introduces an uncomfortable problem for companies with fragmented digital footprints.

Your website says one thing.

LinkedIn says another.

An old directory page lists a previous service.

A 2023 article uses outdated pricing.

A review site uses the old company description.

An old service page remains indexed.

Which version should a system believe?

Microsoft recommends keeping information fresh and maintaining consistency across text, images and media because inconsistent representations create ambiguity.

This is especially important for facts such as:

Company name
Product names
Service scope
Location
Pricing
Availability
Features
Leadership
Contact information
Opening hours
Policies
Specifications

AI search optimization therefore includes information governance.

Not simply content creation.

Step 9: Understand what structured data can—and cannot—do

Schema markup remains useful.

It helps search systems interpret structured information about entities such as organizations, products, articles, events and breadcrumbs.

But there is an important distinction between:

structured data is useful

and:

structured data makes AI cite you.

Google explicitly says structured data is not required specifically for generative AI Search and that there is no special schema.org markup companies need to add to appear in its AI experiences.

So use schema properly.

Use Organization schema to represent a genuine organization.

Use Article or BlogPosting for actual editorial pages.

Use Product for genuine products.

Use BreadcrumbList for real hierarchy.

Do not add five unnecessary schema types because somebody calls it a “GEO stack.”

Schema should clarify the content that exists.

It should not invent meaning that does not.

Step 10: Measure AI search visibility as a competitive system

AI Search Visibility requires repeatable measurement across citations, mentions, recommendations and competitors.

You cannot improve AI visibility intelligently if your measurement consists of occasionally asking ChatGPT:

“Who are the best companies for X?”

and taking a screenshot.

The same question can produce different answers.

Different platforms use different sources.

Market, language, timing and phrasing can change results.

You need a repeatable measurement framework.

Microsoft now provides first-party AI visibility information through Bing Webmaster Tools AI Performance, including total citations, cited pages and grounding queries across supported AI experiences. Microsoft is careful to note that citation frequency does not equal ranking, authority or importance.

Google's Search Console now includes dedicated Generative AI performance reports showing impressions and pages appearing in generative AI Search experiences. Google says those reports were expanded to all websites worldwide as of August 31, 2026.

OpenAI automatically includes utm_source=chatgpt.com in ChatGPT Search referral URLs, allowing publishers to identify visits that actually reach their websites.

These are important improvements.

But businesses still need broader competitive measurement.

A useful AI Search Visibility scorecard

Metric	What it tells you
Buyer-question coverage	How many commercially relevant questions produce your brand?
Brand mention rate	How often are you explicitly named?
Recommendation rate	How often are you actually presented as an option?
Citation rate	How often is your website used as a source?
AI share of voice	How does your presence compare with competitors?
Source influence	Which domains repeatedly shape the answers?
Topic coverage	Which categories do AI systems associate with you?
Referral traffic	How many users actually click through from AI systems?
Commercial outcome	Do those visits or AI-assisted journeys produce enquiries, opportunities or revenue?

The last line matters.

Visibility is a marketing metric.

Commercial impact is the business metric.

A citation is not the same as a recommendation

A website can contribute useful information to an AI answer without its business being recommended.

This deserves repeating because it changes how AI visibility should be managed.

Suppose an accounting firm publishes the best article in its market on:

“How much does outsourced accounting cost in Singapore?”

An AI answer cites that article.

Then recommends three competing accounting firms.

From a content perspective, the firm won.

From a commercial perspective, it may have helped somebody else close the buyer.

The right diagnostic questions are:

Why was our information good enough to cite?

and:

Why was our business not good enough to recommend?

Possible reasons might include:

The commercial service pages are weak
The company is poorly described across the web
Competitors have stronger reviews or evidence
Independent sources repeatedly feature competitors
The business's specialization is unclear
The AI associates the company primarily with educational content rather than the actual service
The brand lacks category-level authority

That is a far more useful investigation than celebrating the citation count alone.

Different AI platforms require a common foundation, not identical tactics

It is tempting to create:

ChatGPT SEO

Gemini SEO

Perplexity SEO

Copilot SEO

as though each requires an entirely separate marketing department.

Platform differences do matter.

But the common foundation is substantial.

Google AI Overviews and AI Mode

Google says its AI Search experiences use its Search index, core ranking systems, RAG and query fan-out. Strong traditional SEO and useful original content therefore remain highly relevant.

ChatGPT Search

OpenAI confirms that public websites can appear in ChatGPT Search and recommends allowing OAI-SearchBot for inclusion in summaries and snippets.

Microsoft Copilot and Bing AI experiences

Microsoft provides direct publisher visibility into citations and grounding queries and recommends technical accessibility, clear structure, evidence and freshness.

Other AI systems

Perplexity, Gemini and other systems have their own retrieval, grounding and citation approaches. Exact weighting and source-selection behavior can change and is not fully public.

So instead of chasing undocumented ranking factors for every model, build the strongest common foundation first:

technically accessible + clearly understood + evidence-rich + distinctive + externally corroborated + measurable.

Then test platform-specific gaps.

What people often misunderstand about AI Search Optimization

“We need llms.txt to rank in Google AI”

Google explicitly says it does not use llms.txt for Google Search and that maintaining one neither improves nor harms Google Search visibility. Other services may choose to use such files, so the correct answer is platform-dependent.

“Every paragraph should be two sentences so AI can parse it”

No.

Google explicitly says there is no requirement to break pages into tiny chunks for generative AI Search. Clear structure is useful. Artificial fragmentation is not a universal ranking factor.

“Adding schema makes ChatGPT cite us”

There is no credible evidence that adding schema creates guaranteed ChatGPT citations.

Use structured data because it accurately represents content and entities—not because someone promised a citation.

“We should create hundreds of long-tail pages for every AI prompt”

That can easily become scaled commodity content.

Google specifically warns against creating pages for every conceivable variation simply to manipulate rankings or generative responses.

“We need lots of Reddit mentions”

You need credible external evidence.

Manufacturing fake discussions is not the same thing.

“If we rank first on Google, ChatGPT will recommend us”

No universal rule supports that conclusion.

Strong search visibility can improve discoverability, but recommendation behavior varies across systems and queries.

“If our domain is cited, GEO is working”

Only partly.

Ask whether the brand itself was mentioned and whether it entered the consideration set.

A practical 90-day AI Search Optimization roadmap

Businesses do not need to solve everything at once.

The first quarter should establish visibility, fix obvious blockers and build a repeatable operating model.

Days 1–30: Diagnose

Start by understanding the current reality.

Map commercially relevant buyer questions.

Test representative prompts across the AI systems that matter to your market.

Record:

Your brand presence
Competitor presence
Recommendations
Citations
Repeated source domains
Incorrect information
Missing categories
Technical accessibility issues

Audit the website for crawlability, indexability, entity clarity, content gaps, evidence and outdated information.

The output should be a diagnosis, not a publishing calendar.

Days 31–60: Strengthen

Fix the most commercially important gaps.

That may include:

Improving commercial service pages
Clarifying entity information
Updating outdated facts
Consolidating duplicate pages
Adding missing decision-stage content
Creating original evidence
Improving direct answers
Adding relevant structured data
Strengthening internal links
Making key content accessible
Improving cited-source quality

Do not attempt to fix every prompt simultaneously.

Prioritize areas closest to revenue and buyer consideration.

Days 61–90: Expand and validate

Now expand the external and competitive footprint.

Identify the publications, comparison sites, directories and communities that repeatedly influence category answers.

Pursue legitimate PR, expert contributions, partnerships and evidence-building opportunities.

Retest the same buyer-question set.

Compare:

Before → After

Look for movement in:

query coverage
citations
recommendations
brand mentions
competitor share
cited sources
accuracy

Then repeat.

AI Search Optimization should become an operating loop:

Measure → Diagnose → Improve → Recheck

How do you know what to fix first?

Use the symptom.

What you observe	Likely area to investigate
You do not rank and AI does not mention you	SEO foundation, content, authority
You rank well but are rarely cited	Retrieval fit, information value, content structure
Your pages are cited but your brand is absent	Entity and commercial positioning
Your brand appears but competitors are recommended first	Evidence, reputation, comparison factors, external authority
AI gives outdated information	Freshness, conflicting sources, indexation
AI misunderstands what you sell	Entity clarity and inconsistent positioning
Competitors dominate third-party sources	PR, reviews, partnerships, category presence
Visibility varies wildly by platform	Platform-specific retrieval and source differences

This is why “publish more content” is often a poor first recommendation.

Different symptoms require different interventions.

What should you ask an AI Search Optimization provider?

The market is filling quickly with GEO and AEO services.

Some are excellent.

Some are SEO services with a new acronym.

Some are selling certainty nobody currently has.

Before hiring a provider, ask:

How do you determine which buyer questions matter commercially?

Which AI platforms do you measure?

How do you distinguish mentions, citations and recommendations?

How do you identify why a competitor appears instead of us?

Do you analyze the sources influencing those answers?

How do you separate technical, content, entity and off-site problems?

How do you integrate this with existing SEO?

What first-party platform data do you use?

How do you handle answer variability?

Which parts of your methodology are based on official guidance, research, observation or hypothesis?

What happens after the initial audit?

How do you connect visibility improvements to business outcomes?

Be cautious with anyone promising:

Guaranteed ChatGPT rankings.

There is no stable universal ChatGPT position equivalent to “Google position #1.”

The responsible promise is:

measure the problem, strengthen the evidence and improve the probability of appearing for the right buyer questions.

Not manufacture certainty.

How should businesses think about the value of AI Search Optimization?

The commercial value depends on the role AI plays in the buying process.

Microsoft Clarity analyzed more than 1,200 publisher and news sites and found AI referrals still represented less than 1% of overall traffic in its dataset, despite growing rapidly. But users arriving from AI systems produced significantly higher sign-up rates than organic search visitors within that specific publisher dataset.

That evidence should be interpreted carefully.

It does not prove AI traffic will convert better for every SaaS company, retailer or professional-services business.

But it illustrates an important point.

AI traffic volume is only one part of the commercial value.

A buyer who reaches your website after an AI conversation may already have:

described the problem,

compared alternatives,

asked about price,

investigated risks,

and narrowed the shortlist.

The website visit can therefore happen later in the decision journey.

That makes pre-click visibility more valuable than traditional attribution models may suggest.

When AI Search Optimization may not be your first priority

AI Search Optimization is not automatically the most important marketing investment.

If Google cannot crawl your site, fix technical SEO first.

If your homepage does not explain what you sell, fix positioning.

If your service pages contain no useful information, improve them.

If customers consistently leave poor reviews because the product is weak, GEO is not the solution.

If your category has little evidence of AI-assisted research today, other acquisition channels may deliver faster returns.

The goal is not to chase a trend.

The goal is to meet buyers where meaningful discovery and decision-making are happening.

Frequently Asked Questions

What is AI search engine optimization?

AI search engine optimization is the process of improving a brand's ability to be discovered, understood, retrieved, cited and represented by AI-powered search and answer systems. It combines traditional SEO foundations with buyer-question research, evidence, entity clarity, off-site authority and AI-specific measurement.

Is AI search optimization the same as GEO?

They overlap heavily. GEO, AEO, AI SEO, LLM optimization and AI Search Optimization are competing labels for related work. “AI Search Optimization” is often the clearest umbrella term because it describes the business outcome without requiring the reader to understand another acronym.

Does AI Search Optimization replace SEO?

No. Google explicitly says its generative AI Search experiences continue to rely on core Search ranking and quality systems, making SEO foundations such as crawlability, indexing, useful content and technical quality relevant to AI visibility.

How do I get my website cited by ChatGPT?

There is no guaranteed method. OpenAI recommends allowing OAI-SearchBot so public content can be discovered for ChatGPT Search summaries and snippets. Beyond technical access, publish accurate, useful and distinctive information that is relevant to the questions users ask.

How can I get my brand recommended by AI?

Recommendations cannot be guaranteed. Improve the evidence available about what your company does, who it serves, how it compares, what customers or independent sources say and why it is relevant to the buyer's question. Then measure which prompts and source patterns consistently produce competitors instead.

Does schema improve AI search visibility?

Relevant schema can help machines understand structured information, but it is not a guaranteed AI citation mechanism. Google explicitly says no special structured data is required for its generative AI Search experiences.

How long does AI Search Optimization take?

There is no universal timeline. Technical changes can be implemented quickly, while new crawling, content recognition, external authority and recommendation changes may take longer. The practical approach is to establish a baseline, make prioritized changes and retest over a consistent set of buyer questions.

How do you measure AI Search Optimization?

Measure citations, brand mentions, recommendation presence, buyer-question coverage, competitor share, cited sources, AI referral traffic and commercial outcomes. Google and Microsoft now provide first-party generative-AI visibility data for their own ecosystems, while broader cross-platform measurement requires structured prompt testing.

Being found is only the first step

The temptation with AI Search Optimization is to search for a new technical trick.

A schema type.

A special file.

A paragraph format.

A prompt pattern.

Those things are easier to sell because they are tangible.

The harder work is more fundamental.

Can the system find you?

Does it understand you?

Is there evidence behind what you say?

Do you contribute information worth citing?

Does the wider web support the same story?

And when the buyer finally asks:

“Who should I consider?”

does your company make the answer?

That is the commercial test.

At Soluma, we approach AI Search Visibility by starting with that diagnosis: the buyer questions that matter, the brands currently winning those answers, the sources shaping them and the gaps that explain why your business is present, absent or misrepresented.

From there, the work may involve technical SEO, content, entity clarity, evidence, structured information, digital PR or a combination of them.

Because the objective is not to become “GEO optimized.”

It is to become more discoverable, more credible and more likely to enter the buyer's consideration set when AI becomes part of the decision.

Sources & Further Reading

This article first appeared on soluma.ai.

Talk to Soluma AI

Soluma AI · AI agency and consulting firm · Singapore, Malaysia, Australia, New Zealand, United States, Canada, United Kingdom, UAE