Summary

AEO (answer engine optimization) is search optimization work carried out so that your content appears as a source in Google AI Overviews, AI Mode and chat-based AI answers. Google's official documentation still defines this work as SEO. Official measurement of that visibility, however, is still rolling out: Google's two announcements describe who the report is open to in different terms. That is why we build our reporting on three separate layers.

Key takeaways

• Google's generative AI optimization guide was published on 15 May 2026²; the document was last updated on 10 July 2026¹.
• The Search Console generative AI report was announced on 3 June 2026⁴; it provides impressions, and the announcement does not list a click metric⁴.
• Search Console's generative AI report covers AI Overviews and AI Mode⁵, with a separate report for Discover⁵; Google's two documents describe the rollout differently⁶.
• Microsoft opened the citation report in Bing Webmaster Tools on 10 February 2026⁷; Bing's search share in Turkey is 0.97%¹².
• AI Overviews passed 2.5 billion monthly active users, and AI Mode passed 1 billion monthly users⁶.
• The AI Overviews trigger rate was 6.49% in January 2025 and 15.69% in November 2025¹¹; it peaked at 24.61% in July 2025¹¹.

What is answer engine optimization (AEO)? A definition

Answer engine optimization (AEO) is search optimization work carried out so that your content is shown as a source inside AI answers. The surfaces it covers are Google AI Overviews and AI Mode, along with chat interfaces such as ChatGPT, Gemini, Perplexity and Microsoft Copilot. Google spells out the abbreviations in its own official documentation: AEO means "answer engine optimization", while GEO means "generative engine optimization"¹.

The same document also names the work. From Google's point of view, optimizing for generative AI search is optimizing for the search experience, and it is still SEO¹. That single sentence makes most pitches that sell AEO as a brand-new discipline replacing SEO debatable. So we say it at the top of this page: what is sold here is not new magic, it is the same work on a new surface.

The guide was published on 15 May 2026², and the document was last updated on 10 July 2026¹. The two are frequently confused: what gets called "the July 2026 guidance" is in fact the update date of that same document — a small but reliable indicator of how carefully a source has been read.

Are AEO and GEO the same thing?

In practice the two describe largely the same work; the terms emerged from different communities and Google defines both in the same paragraph. AEO emphasises the answer itself, GEO the text produced by the generative system. The history of the terminology debate is covered in our GEO explainer; the step-by-step implementation side sits in our AEO strategies guide. Our AI SEO article separates the two distinct jobs that the industry's umbrella term actually refers to.

This page is not a repeat of those three articles. What comes after the definition is here: how this visibility is measured today, where that measurement stops, and what an agency can honestly commit to under those conditions. By the end of the page you will know which report you have, which one you do not, and what our team does inside that gap. We do not repeat the entire knowledge layer here; we give as much as you need to make a decision about the service.

Why we offer AEO as a separate service — and why we do not sell "new magic"

AEO is not a separate discipline, it is the same work on a new surface; we still offer it as a separate service. The reason is simple: two things genuinely changed on this surface. First the eligibility requirement, second the scale. The technical infrastructure and content work inside our SEO service stays where it is; the AEO package adds a question map, answer passages and a separate measurement layer on top of it.

The one technical requirement that changed: your site being included in generative AI features

Google's guide adds one item to the technical requirements of classic search: the site must be included in Search generative AI features in Search Console¹. If that setting is off, you are not shown on those surfaces no matter how good the content is. That is why the first item in our audit is not content but the setting itself — and who changed it, and when.

The other direction of that control is written in the document as well. Sites that opt out of these features do not receive traffic and impressions from them; however, this preference is not used as a ranking signal in search results outside generative AI features⁶. In other words, the answer to "will I be penalised if I opt out?" is not a guess, it is the publisher's own announcement.

The second change is scale. According to Google's announcement, AI Overviews passed 2.5 billion monthly active users and AI Mode passed 1 billion monthly users⁶. On a surface that size, how your brand is summarised is no longer only a traffic question; being described incompletely inside an answer nobody clicks is also a measurable loss, and it comes back to your sales team as an objection.

The combined consequence of those two changes is this: AEO is not a campaign, it is a maintenance discipline. The state of the setting, indexing health and content freshness are items to be re-checked every month. On the agency side, that translates into a fixed checklist and the same three verifications repeated in every report.

What we do not sell: lists of secret settings marketed as "AEO technique". We keep the work bounded by what publishers have written, what we can measure, and the real questions in your market. In the proposal, that corresponds to countable items: eligibility audit, question map, answer passages, evidence layer and a monthly measurement report.

Official measurement of AI visibility: Google's two announcements describe the rollout differently

There are two official reports that show a publisher visibility inside AI answers: one at Google, one at Bing. On the Google side, two separate Google documents describe who the report is open to today in two different ways. The Bing report, meanwhile, measures a surface that holds under one percent share in Turkey. This is the most inconvenient truth about the service we sell, and we say it before the proposal stage. The table below places the two side by side against the same criteria.

What Google's report gives, and what it does not

The Search Console generative AI performance report was announced on 3 June 2026⁴. The metrics listed in the announcement are impressions, pages, countries, devices and dates⁴; a click metric is not on that list. The announcement describes not one screen but separate reports for Search and Discover⁴. The scope of the Search report is narrow: AI Overviews and AI Mode only⁵, with a separate report shown for Discover⁵. Google is opening the report to a subset of sites rather than to all sites⁴.

The real uncertainty is in the rollout. On its corporate blog, Google writes that it is first opening the new controls and insights to a subset of site owners in the United Kingdom⁶. The Search Console help document, on the other hand, explains the reasons for not seeing the report in terms of the property rather than the country: the rollout is spread over time, and the site may not have received enough impressions⁵. Neither document counts Turkey as in scope; one limits scope by country, the other does not relate the rollout to country at all.

Whether the report is open in your account can only be seen by looking at Search Console; that is the first item in our audit. Until the screen appears, we do not fill the gap with guesswork: classic Search Console data, movement in branded queries and page-level impression series are what we have. These are not a direct measure of the AI surface but an indirect indicator, and they are labelled as such in the report. What the AI Overviews surface actually is, is explained in our AI Overview explainer; the question here is not the definition but the measurement.

Bing's citation report and what it is worth in Turkey

Microsoft opened AI Performance inside Bing Webmaster Tools on 10 February 2026; the report shows citations of your content in Microsoft Copilot, Bing's AI summaries and selected partner integrations⁷. The caveat applies not to the whole data set but to a single metric: Microsoft writes that grounding queries data is a sample of total citation activity⁷. On the total citation metric there is a separate note: the number does not indicate the placement or presentation of your content within the answer⁷.

On the Turkish side the picture is this: in July 2026 Google held 80.5%, Yandex 17.5% and Bing 0.97% share¹². So Microsoft's brand-new citation report describes a very small surface in this market. The report is not worthless — it is honest and useful — but on its own it cannot build an AEO measurement system for Turkey.

The Bing report is still a gain: it is the first screen giving a publisher data at citation level, and we enable it routinely for brands working in global markets with English content. For a brand aimed at the Turkish market, we keep its weight low.

On the chat interface side, the official document we have is not a measurement document either: OpenAI's developer documentation gives publishers the ability to manage crawler permissions⁸, not a visibility report. The sum of the table is this: there are two official publisher-facing screens; who one of them is open to today is described differently by two Google documents, and the other measures a small surface in our market.

So how do we measure? The three layers of our AEO reporting

We do not pretend to measure what we cannot measure. Our reporting consists of three layers; what each layer gives and what it does not give is stated in writing at the proposal stage. This section is the commercial counterpart of the gap described above.

Layer 1 — first-party data

The first layer is data from your own property: Search Console performance reports, server and redirect logs, conversion movements on the analytics side. The day the generative AI report opens in your account, that is the first place we will look. Until it opens, we read the classic impression and click series, movement in branded queries and direct traffic together.

The most important output of the first month is the baseline picture: which pages are indexed, which queries you appear for, at what volume your brand name is searched. Every claim of progress made without taking that picture is debatable, and we say so plainly in the first meeting.

Layer 2 — third-party visibility score, and its limits

The second layer is third-party tools. Google warns about these tools explicitly: be cautious of tools that promise ranking success or claim to use "internal" Google metrics, because no third-party tool has access to Google's internal ranking or AI systems¹. That is why we write into the contract what the score we use is not.

The metric we track is named the Semrush AI Visibility score; there is no measurable thing called a bare "AI visibility score". At Modaltrans this score went from 0 to 24+ in three months¹³. A single case is not a generalisation and we do not present it as one; the full study sits on the Modaltrans AEO case study page.

A third-party score is an estimate; what makes it useful is its direction and its time series, not its absolute value. We read the score alongside your competitors' series and always enter it into the report with the name of the tool, because changing the tool changes the number.

Layer 3 — manual query tests and records

The third layer is manual query testing: we ask a selected set of questions at set intervals and archive a screenshot of the answer along with its source list. This method gives samples, not coverage, and it is labelled exactly that way on the first page of the report.

None of the three layers on its own produces a single number called "AI visibility". Used together, they answer three questions: are your pages technically eligible, is the direction of visibility upward, and for which questions are you cited. This is the measurement framework an agency can honestly build today; a proposal promising more than this is billing for something it cannot measure.

Our main AEO work

The service runs across four layers: content, technical eligibility, question map and measurement. The headings below match the scope clauses in the contract one to one; which layer is active in which month varies by project.

Content layer: content that is not a commodity

The strongest sentence in the guide is here. Google writes that content people find unique, engaging and useful will likely affect your site's presence in generative AI search over the long term more than all the other recommendations in the guide¹. That is why we put material that cannot be copied at the centre of production: your own measurements, customer data, expert opinion, pricing logic and examples from the field.

In practice this means working with a list of claims rather than a list of topics. For every page we answer the question "which sentence does this page say for the first time"; pages without an answer are removed from the production plan.

Technical eligibility: indexing, snippets and the generative AI setting

On the technical side the list is short but non-negotiable. For a page to be shown in these features it must be indexed and eligible to be shown with a snippet¹; we also verify that the generative AI setting in Search Console is in the included state¹. Robots rules, snippet restrictions and template-driven indexing losses are standard items of this audit.

Technical auditing is not a one-off job. Template changes, a new category page or a release on the developer side can quietly break eligibility. That is why we repeat the audit monthly and deliver the findings in a form the developer team can act on, at file and line level.

Question map and answer passages

The question map is our own method: we extract the real questions that decide the purchase from customer interviews, the sales team, search data and support records. A single clear passage is written for each question; the passage is built so that it makes sense without the rest of the page being read.

This layer has no external source, because it is not a publisher's rule but our own production discipline. If your sector is software, you can see the long-form version of this flow step by step in our AEO content guide for SaaS.

Measurement and reporting

The evidence layer sits under this heading too: next to every answer there is a figure, a date, a source and an author identity. If we have no source to support a sentence, we do not soften the sentence — we remove it. This page itself was written by the same rule; that is why you will not see a single figure in the pricing section.

The last layer is measurement: which questions you are cited for, which pages are quoted, the direction of the score and its counterpart in classic search all come together in one report. No chart with an unclear source appears in the report; where every number comes from and what it does not measure is written down. We prepare the report in a form your marketing team can present to management: verdict first, then measurement, raw data last.

Our AEO process: what happens in the first 90 days

The process consists of five steps, and the first ninety days are a calendar, not a promise of results. The order and naming of the steps are roicool's methodology; only the technical requirement in the first step rests on the publisher's own document.

The first month goes to eligibility and measurement setup: technical audit, setting check, baseline picture and the first version of the question map. The second month is the production month; answer passages and the evidence layer are written. The third month is measurement and correction: we look at which questions you were cited for and update the map.

1. Eligibility audit: indexing, snippets and the generative AI setting
2. Question map: the real questions that decide the purchase
3. Answer layer: a single clear passage for every question
4. Evidence layer: figure, date, source and author identity
5. Measurement: first-party report, third-party score, manual query test

The logic of this flow is as follows: first we become eligible to be shown, then we choose the right questions, then we write the answer and the evidence behind it. Measurement is not at the end but set up from the first week, because it is not possible to discuss progress without recording the starting state. The output of every step is delivered as a separate document.

At the end of ninety days you have the following: an eligibility report, a prioritised question map, published answer passages and the first comparative measurement. That list does not include "we will reach rank X" or "we will get this many citations"; if it did, we could not support it with data.

Throughout the engagement there is a single owner and a meeting every two weeks. The agenda of that meeting is fixed: what was published last period, what the measurement showed, what will be written this period.

llms.txt and "AEO requirements": what we leave out of our package

We do not sell some of the items we frequently see in competing proposals. The most visible example is llms.txt. There is a lot of content presenting this file as a mandatory "AEO requirement"; put two primary sources side by side and the picture becomes clear.

Who was llms.txt written for?

The specification itself says this is a proposal, not a standard: a proposal to standardise on an /llms.txt file to provide information that helps agents use a website. Its author is Jeremy Howard; first published 3 September 2024, with the second version last changed on 10 August 2026⁹.

The adopting side is not search engines either: AI labs publish this file for their own developer documentation⁹. On the Google Search side the ruling is explicit; you do not need to create new machine-readable files, AI text files or markup to appear in Google, because Google Search does not use them¹.

For an item to enter our package it must meet one of two conditions: either it is explicitly requested in a publisher document, or we can see its effect in our own measurement. Work that meets neither does not enter the proposal; if the client wants it, it is done as a separate item with the expectation written down.

Access is a separate matter and should not be confused with llms.txt. Which permissions you give AI crawlers is determined by robots rules and server-side settings; we check that item by item in the technical section of the audit. Which crawler to allow is a commercial decision, not a technical default; we write the decision and its rationale into the report.

The two sources do not contradict each other; their audiences differ. llms.txt is a proposal that may be meaningful on the agent side, not a requirement for Google visibility. The commercial conclusion follows: we do not bill clients for llms.txt. If they want it we add it, and we write into the report why we added it and what it does not solve.

Click data conflicts: that is why we give no guarantees

On the effect of AI summaries on clicks, three serious measurements point in three different directions. The category has not settled either. The chart below shows how the trigger rate changed direction twice within a single year.

The category has not settled: the AI Overviews trigger rate changed direction twice in one year

According to Semrush's study covering more than 10 million keywords, AI Overviews appeared for 6.49% of keywords in January 2025; the rate rose to 24.61% in July 2025¹¹ and fell to 15.69% in November 2025¹¹. This curve can still change, so building a one-year commitment on it would not be honest.

The same study also finds a shift on the intent side: in January, 91.3% of queries triggering an AI Overview were informational, while by October that share had fallen to 57.1%; summaries on commercial and transactional queries increased¹¹. For a service page the consequence is direct: AI summaries are now entering purchase questions.

The conflict is exactly here. Pew Research's measurement is based on the March 2025 browsing records of 900 US adults¹⁰: users who saw an AI summary clicked a traditional result in 8% of visits, while those who did not saw one clicked in 15%¹⁰. Semrush, comparing the same search terms before and after summaries, found that people clicked slightly more¹¹. Google, for its part, says clicks coming from result pages with AI Overviews are of higher quality³.

The planning effect of this volatility is concrete. We do not tie traffic forecasts to the behaviour of a single surface, and we build the content plan so that it works for both classic search and answer surfaces. That way the same page keeps doing its job whether the rate goes up or down.

All three measurements are serious and all three say different things. That is why we give no ranking or citation guarantee in AEO. What we commit to is process, evidence and reporting: we write down what we did, on what date, and what we were able to measure.

In return we expect one thing from you: not to tie the decision to a single metric. Visibility on AI surfaces gains meaning when it is read together with branded searches, direct traffic and the "I saw you there" sentences in sales calls. A reporting setup that demands a single metric will produce the wrong decision in this category.

Who needs AEO today, and who does not

An AEO budget makes sense if your pages can be indexed and shown with a snippet, and if you have content that answers the purchase question clearly; if those two are missing, they come first. The decision table below is a shortened version of the questions we ask in the first meeting.

The real message of the table is not ranking but sequence. For the brands in the first two rows, an AEO budget is early today; the same money spent on clearing technical debt or producing content that answers questions pays off both in classic search and on answer surfaces. The brands in the third and fourth rows are the group that sees the effect of the work fastest.

In Turkey the category does not yet produce commercial search demand

In our own Search Console records, the 15 queries in the AEO and GEO cluster received a total of 182 impressions and zero clicks over ninety days¹⁴. In other words, in Turkey this category is not a matter of capturing existing demand but of creating it. A brand that sets out knowing this works with the right expectation; a brand expecting customers from an "aeo agency" search will be disappointed.

This table is also our filter. We do not start working with a brand whose expectation is "ChatGPT ranking"; we first propose a short audit showing which surface can be measured today. The audit produces two things: a page-level eligibility table and a prioritised question list. The expectation framework becomes clear through those two documents.

The profile that benefits most is also clear: B2B brands with long sales cycles, research-driven decisions and sound technical infrastructure. For those brands, being cited on a single answer surface moves branded search and proposal requests together.

In two of the five situations in the table our answer is no. It may look odd for an agency to write down the cases in which it does not sell its own service; for us this is the section that determines the credibility of the proposal. You will see which box we placed you in, with the reasoning, in the first audit report; if the box changes, we tell you along with the reason.

AEO pricing: why we have no fixed package

We do not publish a fixed AEO package, because two engagements sold under the same price tag turn out to be completely different pieces of work. One brand's technical debt may be clean while another's pages are not being indexed; selling both the same package means over-serving one and under-serving the other. A published list price does nothing except hide that difference.

The four variables that determine the proposal

• Existing content inventory and technical debt: how many pages are eligible, how many need rewriting.
• Number of target languages and markets: a single-language site and a three-language structure are not the same job.
• Measurement access: is the generative AI report visible in the account, which tools do we have access to.
• Production volume: how many answer passages, evidence layers and updates are planned per month.

The pricing conversation starts with those four headings as well. We first run a short audit and determine which layers will be active; if the scope changes the price changes too, and that change arrives in writing as a separate item. The only way to avoid surprise invoices is to write the scope down item by item from the start.

The proposal is written according to the answers to those four headings, and the scope sits in the contract item by item. If you would like to weigh the subject up on your own first, our AEO ebook covers the "is SEO dead" debate and which old tactics no longer work; it does not replace the measurement framework on this page, and it arrives by email once you fill in the form. After that, in an audit call, we look together at the current state of your pages; you leave that call with at least an eligibility table and a first question list.

Frequently Asked Questions

Will AEO replace SEO?

No. Google's official documentation defines optimizing for generative AI search as optimizing for the search experience — that is, still as SEO. What changed is not the discipline but the surface: the answer itself is now a result type. In practice technical eligibility, content quality and the evidence layer are the same team's job; AEO plans and measures them additionally for AI answers.

What exactly does an AEO agency do?

An AEO agency builds the content, technical eligibility and measurement layers together so that your content can be cited as a source in AI answers. For us that means an eligibility audit, a question map, answer passages, an evidence layer and three-layer reporting. We also state in writing what we leave out of scope, because what we do not do counts as part of the proposal too.

How do I measure whether my brand appears in ChatGPT?

As of 15 August 2026, official reports showing a publisher citations or impressions are documented in two places: the Search Console generative AI performance report and Bing Webmaster Tools AI Performance. Neither covers ChatGPT. OpenAI's developer documentation — the only official document we can read — offers crawler permission management, not a visibility report. On the ChatGPT side we are left with three routes: server and redirect logs, manual query testing and a third-party score. All three give samples, not coverage.

Does AEO require special schema.org markup?

No. According to Google, structured data is not required for generative AI search and there is no special schema.org markup you need to add. It still makes sense to keep using structured data as part of your general SEO strategy; it provides eligibility for rich results. We preserve and validate existing markup rather than producing new markup in the name of AEO.

How long does AEO work take to produce results?

There is no time frame that can be guaranteed. In the clearest example we have measured, a Semrush AI Visibility score went from 0 to 24+ in three months; this is a roicool measurement and cannot be independently verified, with the detail on the Modaltrans AEO case study page. A single case is not a generalisation; the outcome varies with sector, existing content inventory and technical debt. That is why we give a calendar, not a results date.

If I opt my site out of AI answers, will my rankings drop?

No. Google writes that this control will not be used as a ranking signal in search results outside generative AI features. But there is a cost: sites that opt out receive no traffic and no impressions from those features. So your classic rankings are preserved while your visibility on the AI surface ends. Before deciding, we recommend measuring the impressions coming from that surface.

Sources

1. Google Search Central. Optimizing your website for generative AI features on Google Search. https://developers.google.com/search/docs/fundamentals/ai-optimization (2026)
2. Google Search Central Blog. A new resource for optimizing for generative AI in Google Search. https://developers.google.com/search/blog/2026/05/a-new-resource-for-optimizing (2026)
3. Google Search Central. AI features and your website. https://developers.google.com/search/docs/appearance/ai-features (2025)
4. Google Search Central Blog. Introducing Search Generative AI performance reports in Search Console. https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports (2026)
5. Google Search Console Help. Generative AI performance report (Search) — Search Console Help. https://support.google.com/webmasters/answer/16984139?hl=en (2026)
6. Google (The Keyword). New opportunities, control and insights for website owners. https://blog.google/products-and-platforms/products/search/new-controls (2026)
7. Microsoft — Bing Webmaster Blog. Introducing AI Performance in Bing Webmaster Tools Public Preview. https://blogs.bing.com/webmaster/february-2026/Introducing-AI-Performance (2026)
8. OpenAI developer documentation. Overview of OpenAI Crawlers. https://developers.openai.com/api/docs/bots (2026)
9. llmstxt.org — Jeremy Howard. The /llms.txt file, v2. https://llmstxt.org/ (2026)
10. Pew Research Center. Google users are less likely to click on links when an AI summary appears in the results. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click (2025)
11. Semrush. Semrush AI Overviews Study: What 2025 SEO Data Tells Us About Google's Search Shift. https://www.semrush.com/blog/semrush-ai-overviews-study/ (2025)
12. StatCounter GlobalStats. Search Engine Market Share in Turkey — July 2026. https://gs.statcounter.com/search-engine-market-share/all/turkey (2026)
13. roicool. Modaltrans AEO study — Semrush AI Visibility score measurement. roicool measurement — case page published, cannot be independently verified (2026)
14. roicool. roicool Search Console records, 1 May – 29 July 2026. roicool internal data — cannot be independently verified (2026)

‍