Generative engine optimization (GEO): be one of the sources the AI answer is built from
Your pages used as sources when Google's AI Overviews and AI Mode, Gemini, and Copilot generate an answer to a buyer's question, with a monthly record of which prompts include you, which include a competitor, and what changed.
For companies whose Google traffic has fallen while impressions held, and for anyone who has searched their own category and found an AI summary sitting above the results they used to win.
Updated
What is generative engine optimization (GEO)?
Generative engine optimization (GEO) is the practice of making a website more likely to be retrieved, used, and cited when a generative engine composes an answer instead of listing links. The generative engines that matter for most companies are Google's AI Overviews and AI Mode, Gemini, Microsoft Copilot, and the search modes of ChatGPT and Perplexity. The term comes from a 2023 research paper that measured which content changes raised a source's visibility in generated answers; the findings that held up were adding citations and statistics, quoting named sources, and writing plainly. In practice GEO overlaps almost entirely with answer engine optimization and with well-executed SEO. What is specific to GEO is the mechanism: generative engines run several related searches for one prompt (query fan-out), pull passages from the top results of each, and synthesise them. So a page has to rank for the sub-questions, not only the head term, and each passage has to stand alone. At Seorythm the GEO service covers a fan-out audit, passage restructuring, freshness and provenance signals, and monthly tracking of AI Overview and AI Mode inclusion for a fixed prompt set.
What you get
| Deliverable | Cadence | What it includes |
|---|---|---|
| Fan-out and inclusion audit | Weeks 1 to 2 | 40 to 80 buyer prompts run across AI Overviews, AI Mode, Gemini, and Copilot. For each, the sub-questions the engine expands to, the pages it cites, whether you appear, and which competitor does. Search Console queries with Overviews present, and their click trend. |
| Coverage plan | Month 1 | Which sub-questions your existing pages already answer, which need a new section, and which need a new page. Mapped to the prompts that matter, so nothing is written for a question nobody asks. |
| Passage restructuring | 4 to 8 pages per month | Question-led headings, a direct answer in the first sentence, a figure or definition in the passage, source named in text. Marked with speakable structured data where it fits. |
| Freshness and provenance signals | Month 1, then ongoing | Visible and structured dateModified, named author with credentials, Organization data, and a published review cadence for the pages that carry the answers. |
| Statistics and citable claims | Quarterly | One piece of original data or a documented method per quarter that other pages can cite, because generated answers favour sources with specific numbers and named origins. |
| Inclusion tracking and reporting | Monthly | The prompt set re-run and sampled, inclusion share per engine, competitor share, Overview-present queries in Search Console with impressions and clicks, and the pages that gained or lost inclusion. |
How the work is done
Map the fan-out before writing anything
When someone types a question into AI Mode, Google does not run one search. It runs several, one for each sub-question it thinks the prompt implies, and builds the answer from the top results of each. Google calls this query fan-out, and it changes what a page has to do. A single page ranking for the head term is no longer enough; the sources the engine cites are the pages that rank for the pieces. So the audit starts by running each of your 40 to 80 buyer prompts, recording the sub-questions the engine expands to, and listing which pages it cited for each. That map tells us what to write and, as often, what not to.
Rank for the sub-question, not only the head term
Google’s own guidance on appearing in AI features says there is nothing special to do beyond the fundamentals, and for retrieval that is true: AI Overviews pull from the same index the ten blue links come from. The GEO difference is coverage. Where the fan-out map shows a sub-question your site does not answer, we add a section or a page, following the same on-page SEO discipline as any other page. Where you already rank for the sub-question but are not cited, the problem is usually the passage, which is the next step.
Passages with a number in them
The original GEO research, a 2023 paper that tested which changes raised a source’s visibility in generated answers, found that adding statistics, quoting named sources, and including citations moved the needle, while keyword stuffing did nothing or hurt. That matches what we see in the sampling. A passage gets used when it answers the question in its first sentence, contains a specific figure or definition, names where the figure came from, and makes sense when lifted out of the page. We rewrite the key passages to that pattern and mark them with speakable structured data. Once a quarter we produce one piece of original data or a documented method, because the pages that get cited most are the ones other pages cite.
Show the date and the author
Generated answers prefer recent, attributed sources, and they check. Every page carrying an answer gets a visible and structured dateModified, a named author with a credentials page, and Organization markup that agrees with your profiles elsewhere. This is technical SEO housekeeping with a specific purpose: when two passages say the same thing, the engine tends to use the one it can date and attribute.
Measure inclusion and clicks, separately
There is no rank for an AI Overview, so we build the measurement. The prompt set is re-run and sampled monthly across AI Overviews, AI Mode, Gemini, and Copilot, and we report inclusion share per engine, next to competitor share. Alongside that we watch Search Console for queries where an Overview is present and track their impressions and clicks over time, because being cited and being clicked are different outcomes, and for some purely informational queries the second never follows the first. Where that is the case we say so and move the budget to questions where the answer leads somewhere.
GEO is one half of our AI search programme; the other is answer engine optimization, which covers the conversational assistants. Both sit on top of the SEO services work that makes a page retrievable in the first place. If you want the terms untangled, read SEO vs AEO vs GEO vs AIO vs LLMO vs SXO. To find out where you stand, start with the audit, or see pricing for the standalone programme.
What it costs
Included in the sprint or retainer, or a standalone AI search programme
The fan-out and inclusion audit is part of the $2,800 audit. Ongoing GEO work runs inside the sprint or retainer, or as a standalone AI search programme at $1,200 to $4,000 a month, the same programme and price as our AEO service, because the work overlaps. The comparison is on the pricing page.
What moves the number
- Number of prompts tracked and engines covered
- How many pages need new sections versus edits
- Whether original data must be produced to be citable
- Whether the site already ranks for the sub-questions or needs SEO first
Who this is not for
- Sites that do not rank on page one for the underlying queries. Generative engines retrieve from the index; GEO cannot place a page the engine never sees. Start with SEO services.
- Anyone who wants pages written to trigger an Overview for a query with no commercial value. Inclusion is not the goal; the buyer who reads the answer is.
- Buyers who expect a guaranteed appearance for a given prompt. Overviews change between sessions and locations, and nobody can honestly promise one.
- Companies unwilling to publish real facts, prices, or data. Generated answers favour specific claims with a named source; vague pages are passed over.
Risks and how we manage them
| Risk | How we manage it |
|---|---|
| AI Overviews take the click even when you are cited. | We track clicks on Overview-present queries separately and prioritise questions where the answer leads to a next step (a comparison, a price, a tool) that the summary cannot complete. Where a query is purely informational and the Overview satisfies it, we say so and stop spending on it. |
| Google changes how Overviews retrieve or display sources. | The work rests on inputs that every retrieval system rewards (ranking, plain passages, specific claims, provenance) rather than on a display quirk. Monthly tracking shows a change within one cycle. |
| Optimising for prompts nobody asks. | The prompt set is built from your sales calls, support tickets, and Search Console, then pruned quarterly against real impressions. |
| Reading noise as movement. | Each prompt is sampled several times per engine per month. We report inclusion share and label changes inside the noise band as such. |
How is GEO different from SEO and AEO?
SEO earns a position in a list of links. AEO earns a citation in a conversational answer from ChatGPT, Perplexity, or Claude. GEO sits between the two and is mostly about Google: AI Overviews and AI Mode retrieve from Google's own index, so ranking still matters, but they retrieve for the several sub-questions a prompt implies, and they quote passages rather than pages. The practical differences are three. First, coverage: a GEO plan targets the cluster of related questions the engine fans out to, not one head keyword. Second, unit of content: a self-contained passage that is true when lifted out of the page, with a specific figure or definition in it. Third, measurement: there is no rank for an AI Overview, so we sample prompts and record inclusion share, and we watch the click data in Search Console for the queries where an Overview appears. A site that already ranks and is written plainly is most of the way there. A site that does not rank cannot be fixed with GEO alone, because the engine never retrieves it.
Questions buyers ask
What does a GEO agency do?
It finds the questions a generative engine answers in your category, works out which sources the engine is drawing on and why, and then makes your pages the easier source to use, by ranking for the sub-questions, writing passages that can be quoted whole, publishing specific and attributable facts, and keeping dates and authorship visible. It measures the result by sampling prompts every month, because there is no rank tracker for a generated answer.
Is GEO the same as AEO?
Almost. Both aim to be cited in an AI-composed answer. GEO is the term most used for Google's AI Overviews and AI Mode and for the research lineage that studied generated answers; AEO is more often used for conversational assistants like ChatGPT and Perplexity. The inputs are the same, so we sell them as one AI search programme and keep two pages because buyers search for both terms and the engines behave a little differently.
How much does GEO cost?
The fan-out and inclusion audit is included in our $2,800 audit. Ongoing work runs inside the sprint ($4,800 to $7,400 a month) or the retainer ($2,300 to $4,900), or as a standalone AI search programme at $1,200 to $4,000 a month, set 20% below the closest published benchmark. The comparison is on the pricing page.
Does GEO replace SEO?
No. Google's generative features retrieve from Google's index and lean heavily on pages that already rank. GEO adds the coverage, passage structure, and measurement layer on top of a site that ranks; it cannot substitute for one that does not.
How long does GEO take to show results?
Inclusion for prompts where you already rank can change within four to eight weeks of restructuring, because the engine re-fetches ranking pages often. Prompts that need new pages or new rankings take as long as SEO takes for those pages, typically three to six months.
Do you use AI to write the GEO content?
For research, fan-out mapping, prompt sampling, and QA, yes. Passages are written or rewritten by a named person and reviewed by your expert, because a generated answer that quotes a generic passage rarely names its source.
Start with the audit.
Two weeks, fixed price, and a prioritised list you can act on with or without us. If SEO is not the right channel for you, the audit will say so.