Generative engine optimization services, measured before and after.
A buyer asks ChatGPT for the best tool in your category and gets a competitor, sometimes with a citation to your own comparison page. This page shows how to measure how often that happens across four engines, fix the pages that cause it, and run the same measurement again. Inside a Growth Pilot that is week one's baseline and two experiment cards. The number is the deliverable.
By Daria Dovzhikova · Updated October 2026
The method, in five steps
A baseline, the fixes the baseline names, and the same test again.
The baseline
Forty-plus unbranded buyer questions in your category, run against ChatGPT, Claude, Gemini and Perplexity. Branded prompts run separately and sit on their own line. You get the four numbers per engine before anything changes.
Extraction fixes
The pages the engines already open and misread: comparison pages that feed a rival's answer, pricing pages with the answer buried, docs with no extractable summary. Each one rewritten so the machine pulls your claim, not the other company's.
Canonical positioning
One positioning statement, one set of capability claims, one product description, carried identically across llms.txt, JSON-LD, answer-first page summaries and the GitHub or listing surfaces the engines cite. Engines repeat what they see repeated.
Comparison and alternative pages
Sourced decision guides for the questions buyers actually ask an engine: alternatives to the incumbent, migration from the tool they are leaving, you versus the two rivals named instead of you. Built to be cited, with the citations pointing home.
The re-measurement
The same forty-plus prompts, the same four engines, the same scoring. A one-page readout with the delta per engine, the pages that moved it, and the supply-chain list for the next placements. If a number did not move, the readout says so.
Inside a Growth Pilot
The baseline is week one of the pilot. The fixes it names become the two experiment cards, decided every Friday and re-measured on day 28. $5,000 fixed, four weeks. Continuation at $2,500 a month keeps the re-measurement running after.
How it is measured
Four numbers. None of them blended.
Most AI visibility dashboards ask the engine about your brand by name, count the yes, and call it visibility. That measures whether the model can repeat a word you just gave it. The number below that matters is the one where nobody said your name.
Organic citation share, per engine
Of the unbranded buyer questions where any tracked vendor is named, how often is it you? Scored per engine, never blended. The same brand can be first on Perplexity and absent from Gemini in the same week, because the engines draw on different sources.
Branded recall, on its own line
Does the engine know you when handed your name? Almost everyone scores high here. A dashboard that averages this into the visibility number is reporting recall as discovery, which is how a company with zero organic presence ends up with a 95% score.
Named instead of you
The vendors the engines surface on your organic questions. This is your competitive set in the AI era, and it is often not the set on your battle cards.
The citation supply chain
The domains the engines cite when they answer. Engines cite what other people publish about you, so this list is the placement roadmap: which listicles, communities and partner docs need to carry your name.
The method and the scorer are public: the write-up is at Your AI visibility number is lying to you, and the open-source scorer is at github.com/dovzhikova/ai-visibility-scorer. You can rerun the baseline yourself, before or after hiring anyone.
What a GEO agency retainer gets you, next to a pilot
An AEO agency or GEO agency sells the same thing an SEO agency sold in 2019: a monthly retainer, a content quota, and a score. The pricing guides put mid-market retainers at $2,000 to $8,000 a month and enterprise tiers at $10,000 to $25,000 and up, with no defined end date. The pilot below has an end date, and the end date has a number attached.
| Axis | Typical GEO agency retainer | Growth Pilot |
|---|---|---|
| Price | $2,000 to $8,000 a month for mid-market, $10,000 to $25,000+ for enterprise tiers, open-ended | $5,000 fixed, four weeks, as the Growth Pilot; Continuation at $2,500 a month after, optional |
| Scope | A monthly content quota plus monitoring | One baseline, the fixes the baseline names as two experiment cards, one re-measurement |
| The number reported | A blended visibility score, usually with branded prompts mixed in | Four numbers per engine, branded recall kept on its own line |
| What gets fixed first | New content, because the quota needs filling | The pages the engines already cite and misread, because that is where the delta is |
| Guarantee | Some promise placement. Nobody controls the model. | No placement guarantee. A measurement guarantee: the same test before and after, published to you |
| What you keep | The content | The content, the prompt set, the scoring method, the supply-chain list, and an open-source scorer to rerun it yourself |
Retainer ranges from Digital Elevator's 2026 AEO and GEO pricing guide and WebFX's 2026 GEO cost breakdown. Both are agency-published; read them as the category describing itself.
Measured on a real product
Radar by Skyhook: 3 of 48 to 8 of 46 in eight weeks.
Unbranded answers naming Radar across four engines. The incumbent was named in 39.
Same prompt design, same engines. First Google AI snippet for the product arrived mid-way.
The project's own comparison page was the page engines cited when recommending a competitor. Rewritten as a sourced decision guide.
Radar is an open-source Kubernetes UI. The engines had read its Lens-alternative page and concluded the alternative was someone else. Fixing that page, applying one canonical positioning across llms.txt, JSON-LD and the blog summaries, and adding answer-first sections did the rest. Nothing in that list is a trick. It is reading what the engine reads, then correcting it.
What this will not do
- //Guarantee a placement. Nobody controls what a model says. A vendor promising a ranking inside ChatGPT is selling you a screenshot.
- //Sell you llms.txt as the lever. Ahrefs pulled server logs across 137,000 domains and found 97% of llms.txt files received zero requests in a month. SE Ranking modeled 300,000 domains and found the file added noise to citation prediction, not signal. A pilot ships one because it is cheap and harmless. It is not where the delta comes from.
- //Report one blended score. If you want a single number that goes up, several dashboards will sell you one. It will go up because it counts recall.
- //Write forty blog posts. The baseline usually says the pages you already have are the problem, and rewriting six of them beats publishing forty new ones.
Studies cited: Ahrefs, 137,000 domains, May 2026 and SE Ranking, 300,000 domains.
Built for
A developer-tool or AI-product company with a category question.
A founder who has typed the category question into ChatGPT, seen a rival, and wants a number for how often that happens. Inside a Growth Pilot this is one of the two experiment cards, or both.
Especially useful when a competitor is running migration content against you, or when your own comparison pages are older than the models reading them.
Not built for
- //Pre-product companies. There is no category question yet, so there is nothing to measure.
- //Local businesses and consumer brands. The four engines cite different sources there, and the method is tuned for B2B software.
- //Anyone who wants a single score that goes up. Several dashboards sell that. It goes up because it counts recall.
Pricing
- $5,000 fixed for the four-week Growth Pilot, half at kickoff and half on the last day
- $2,500 a month for Continuation after, optional, month to month
- Every price and term is on the pricing page
Generative engine optimization services: questions buyers ask
How much do generative engine optimization services cost?
As a Growth Pilot, $5,000 fixed for four weeks, half at kickoff and half on the last day: the baseline in week one, the fixes as two experiment cards decided every Friday, the re-measurement on day 28. Continuation at $2,500 a month keeps the monthly re-run going after. For context, mid-market AEO and GEO retainers run about $2,000 to $8,000 a month with no defined end, per Digital Elevator's 2026 pricing guide, and WebFX puts medium-business AI search retainers at $5,000 to $25,000 and up a month.
How long does it take to see results?
Four weeks to the day-28 readout. On the Radar engagement the baseline moved from 3 of 48 unbranded answers to 8 of 46 in eight weeks, with the first Google AI snippet arriving mid-way. Engines re-crawl and re-weight on their own schedules, so some fixes register in days and some in a month. That is why the Friday reviews decide continue, change or stop, and why Continuation exists for the fixes that need longer.
What exactly do I get?
A baseline report with the four numbers per engine, the rewritten extraction pages, the canonical positioning applied across llms.txt, JSON-LD and page summaries, two to four sourced comparison or alternatives pages, a re-measurement readout with the delta per engine, the supply-chain list for the next placements, and the prompt set plus scoring method so you can rerun it.
Which engines are covered?
ChatGPT with search, Claude, Gemini and Perplexity, scored separately. Google AI Overviews are read where they appear on the same queries. Each engine gets its own line because they cite different sources and move at different speeds.
How is it measured?
The same set of unbranded buyer questions, run against the same engines, before and after. Organic citation share is counted only where no vendor name was in the prompt. Branded recall is measured with your name in the prompt and reported separately. The method is public: the scorer is open source, and the write-up is on the GTM Labs Substack.
How is this different from an SEO agency or an AEO agency?
An SEO agency optimizes for a ranking on a results page. An AEO agency or GEO agency usually sells a monthly retainer and reports a blended score. This page treats AI visibility as a measurement problem first: find the pages the engines already cite and misread, fix those, then measure again with branded prompts kept out of the number. Fixed scope, fixed price, one readout.
Do you guarantee results?
No placement guarantee, because nobody controls the model. The guarantee is the measurement: the same test before and after, with the delta per engine reported to you whether it moved or not. The Radar readout would have said 3 of 48 to 3 of 46 if that had been the outcome.
What happens after the pilot?
Three options. Stop, with everything handed over and rerunnable. Continuation at $2,500 a month, which reruns the baseline monthly, ships the fixes the re-run names as one new card a month, and reports on one page. Or the supply-chain list becomes a placement project, which is a separate scope.
What is GEO in digital marketing?
Generative engine optimization is the work of getting a brand named and cited in the answers that ChatGPT, Perplexity, Gemini and Google's AI surfaces generate. Classic SEO earns a ranking on a results page. GEO earns a place inside the answer, where there is no page two. The inputs overlap with SEO, clear pages and real third-party mentions, but the measurement is different: not position, but the share of unbranded buyer questions where the engine names you.
What is the difference between SEO, AEO and GEO?
SEO gets a page ranked in a list of links. AEO, answer engine optimization, gets a page picked as the answer to a direct question, which used to mean featured snippets and now includes Google's AI Overviews and AI Mode. GEO gets a brand named or cited inside a generated answer from ChatGPT, Perplexity, Gemini or Claude. In 2026 AEO and GEO describe the same job with different vendors' vocabulary. I measure one thing across all of it: whether the engine reaches for you on a question that does not contain your name.
What is LLM SEO?
Another name for GEO, used mostly by SEO teams and the tools sold to them. LLM SEO means optimising for large language models: the pages they can extract from, the sources they trust, and the entities they associate with a category. The method on this page is LLM SEO by that definition. The only part of the name I would drop is the promise of ranking; a model does not rank you, it either reaches for you or it does not.
Related guides
Want the baseline before you decide?
Thirty minutes to scope it. If your category question already returns you, the call says so and there is no pilot to sell.