TL;DR
- What it is: Making your pages easy for AI engines to retrieve and cite. For Google, it's still SEO.
- How a page gets picked: The engine writes sub-queries, retrieves pages, and cites about half. ChatGPT cited 88% of the URLs that came from search.
- Google's view: Write what only you can write. Skip llms.txt, chunking and special schema.
- Five tactics, graded: Only some have real evidence. The table says which.
- Measure it: Search Console has a Generative AI report. For the other engines, run your own prompts.
Does AI name your brand? Check your site.
AI search optimization is three jobs.
- Get indexed.
- Answer the sub-questions the engine asks behind the scenes.
- Get named on the pages it reads.
That's it. And for Google Search, Google says this is still SEO.
How big is this channel? Two numbers.
- Up 693%. Traffic to US retail sites from generative AI tools, year over year, in the 2025 holiday season. (Adobe.) A growth rate. Not a share of traffic.
- 37 of 5,375. The visitors ChatGPT, Claude and Perplexity sent to our own site. 1 July to 30 September 2026.
Growing fast. Still small.
Most advice on this topic is stated as fact. So I graded five tactics by how much evidence each one has.
What AI search optimization is and is not
The work of being found, understood and cited when someone asks an AI tool a question.
The tools: ChatGPT, Perplexity, Gemini, Copilot, and Google's AI Overviews and AI Mode.
It has other names. Answer engine optimization (AEO). Generative engine optimization (GEO). LLM optimization (LLMO). Same work, different angles. This comparison explains the overlap.
Google says SEO still applies. Its guide says:
"From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
What it is not:
- A fix for thin content. The guide says unique, helpful content will likely influence your presence in AI search more than any other suggestion in it.
- A separate channel on Google. Its AI features pull from the Search index and core ranking systems. (Other engines retrieve differently. See AI search engine optimization.)
- Visible in the usual reports. A mention with no click leaves no visit in your analytics.
How an AI answer picks your page
Google describes its AI features with two terms.
- Retrieval-augmented generation (RAG). The system pulls relevant pages from the Search index. It writes an answer from them, with links.
- Query fan-out. The model writes related queries and runs them at once.
Google's example. For "how to fix a lawn that's full of weeds", the fan-out queries might include:
- "best herbicides for lawns"
- "remove weeds without chemicals"
- "how to prevent weeds in lawn"
ChatGPT shows the same shape.
Ahrefs studied 1.4 million ChatGPT prompts (April 2026). ChatGPT retrieves dozens of URLs per prompt. It cites about half.
An average prompt had about 16.6 cited and 16.6 non-cited URLs.
Four findings stand out:
- Search URLs get cited most. 88% of the URLs that came from search were cited. For news URLs it was 12%.
- Reddit is read, not credited. It made up 67.8% of the non-cited URLs. It was cited 1.93% of the time.
- Titles come first. Each result comes back with a title, a snippet and a URL before ChatGPT opens the page. Cited titles were closer in meaning to the prompt (0.602 against 0.484). And closer still to the best-matching fan-out query (0.656). (Ahrefs used open-source embeddings. An approximation.)
- Plain URLs went with more citations. Results with natural-language slugs were cited 89.78% of the time. Against 81.11% without.
So do you need to rank? Two studies. Two directions.
- Ahrefs writes that your content needs to rank.
- Semrush (July 2025, 500+ digital marketing and SEO topics) found that ChatGPT search cited pages in traditional positions 21 and below, for related queries, almost 90% of the time. It also found that some LLMs cite positions 1 to 5 more often than 6 to 20.
One reading: a page can rank low for the prompt as typed. And high for one of its sub-questions.
Neither study tests that. So it's a reading. Not a finding.
The working rule: be indexed. Be the best answer to a sub-question. Don't treat a rank number as the entry ticket.
Treat both studies with care. Semrush's covers one field. Ahrefs says its 50% covers the whole path from retrieval to citation, and some non-cited URLs were likely never opened.
What Google says to do and to skip
Google published its guide on 15 May 2026. Last updated 10 July 2026.
It covers Google Search only. Not ChatGPT or Perplexity.
What Google says to do
Write non-commodity content.
- Google's commodity example: "7 Tips for First-Time Homebuyers".
- Its non-commodity example: "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line".
A first-hand review gives a unique point of view. A summary only restates what exists.
Keep the technical basics. A page must be indexed and eligible to show in Google Search with a snippet. Content must be crawlable. (Google adds that indexing and serving are not guaranteed.)
Check the Search generative AI control. In Search Console, your site must be included in Search generative AI features. Include is the default. Excluded? You get no links, impressions or traffic from AI Overviews, AI Mode or the Discover AI features. A change takes a few days.
Add images, video and business details. The AI features can show images and video that support the text. Merchant Center feeds and Google Business Profile can make products and services visible in AI responses.
What Google says to skip
- llms.txt and other special files or markup. Google Search doesn't use them. Keeping one will neither harm nor help your Google visibility. It's fine to keep one for other services.
- Chunking. Google: "There's no requirement to break your content into tiny pieces for AI to better understand it."
- Rewriting content just for AI. AI systems understand synonyms and meaning. And a separate page for every fan-out query, made mainly to manipulate results, violates its scaled content abuse policy.
- Inauthentic mentions. Google says seeking them "isn't as helpful as it might seem".
- Special schema. Structured data is not required for generative AI search. No special schema.org markup is needed. (It still helps you qualify for rich results.)
Google speaks only for Google. For ChatGPT, see tactic 3 below.
Five tactics and how much evidence each has
Most advice on this topic is stated as fact.
Google's own test for third-party advice fits here. Good advice either qualifies its claims as opinion based on data or experience, or cites Google's guidance.
So each tactic gets a grade:
- Documented: Google or OpenAI says it.
- Measured: a study found it.
- Opinion: one analyst, one data point, or reasoning without a test.
| Tactic | What the source says | Grade |
|---|---|---|
| 1. Be in the search pool for the sub-questions | Ahrefs: ChatGPT cited 88% of the URLs that came from search. Semrush: cited pages often ranked 21 or lower. Google: its AI features rest on core ranking systems | Documented (Google). Measured (two studies) |
| 2. Say something only you can say | Google: unique, non-commodity content will likely matter more than any other suggestion in its guide. No data given | Documented, no data |
| 3. Let the right bots in | OpenAI: OAI-SearchBot decides whether you show in ChatGPT search. Google: a page must be indexed, snippet-eligible and included | Documented (OpenAI and Google) |
| 4. Answer first, name specifics, match the sub-questions | Ahrefs: cited titles sit closer to the prompt and the fan-out queries. Kevin Indig: 44.2% of citations came from the first 30% of a page | Measured, correlation only |
| 5. Get named on pages the engines read | Our scan for "best SEO tool": 14 of 24 cited pages were listicles. Google: fake mentions are not as helpful as they seem | Opinion, plus one data point |
1. Be in the search pool
- Pick 10 to 20 questions your buyers ask.
- Check that a page of yours for each one is indexed.
- Write the follow-up questions a buyer would ask next. Those are your sub-questions.
- Make sure a page answers each one.
Your list is a stand-in for the engine's real sub-queries.
Don't use a rank number as a pass mark. Neither study says one is required.
2. Say something only you can say
Add what only you have. Your own data. Tests. Screenshots. Customer cases.
Google gives no data for this one.
It says only that unique content will likely matter most in the long run. And that a summary of existing content restates what's already available elsewhere.
3. Let the right bots in
OpenAI documents four bots. Three matter here:
- OAI-SearchBot surfaces sites in ChatGPT search results.
- GPTBot crawls content that may be used for training.
- ChatGPT-User visits a page when a user asks. robots.txt rules may not apply.
The settings are independent. You can allow OAI-SearchBot and disallow GPTBot.
A site that opts out of OAI-SearchBot won't be shown in ChatGPT search answers.
OpenAI says a robots.txt change can take about 24 hours to take effect for search.
User-agent: OAI-SearchBot
Allow: /
Then check the Search generative AI control from the section above.
4. Answer first, name specifics
Kevin Indig's analysis of 1.2 million search results reports:
- 44.2% of citations came from the first 30% of a page.
- Cited content had an entity density of 20.6%.
- It used more definitive language. 36.2% against 20.2%.
One analyst's correlation. And the full post is paid. Check his method before you rely on it.
Put the answer first anyway. The reader wants it first.
An illustration, not a claim:
Before: Many site owners wonder how long a robots.txt change takes to matter, and the answer can vary depending on several factors.
After: OpenAI says ChatGPT search can take about 24 hours to adjust to a robots.txt change.
Don't fake certainty. A claim you can't support should be labeled as opinion.
5. Get named on pages the engines read
- Ask your category question in each engine.
- Read the listicles and comparison pages it cites.
- Pitch those authors with something true and specific.
The method: how to get into AI visibility listicles.
On Reddit, answer real questions. But don't expect ChatGPT to cite the thread. Ahrefs found a 1.93% citation rate.
A real example of what gets cited
Our leaderboard scans "best X" questions on five AI engines.
Its best SEO tool page, dated 24 September 2026, lists 24 cited pages:
| Page type | Cited pages |
|---|---|
| Listicles | 14 |
| Comparisons | 3 |
| Vendor page ("All Ahrefs tools") | 1 |
| Other types | 6 |
14 of 24 are lists. One is a vendor page.
Ahrefs and Semrush were the first names. Average ranks of 1.9 and 2.1 across engines.
Read this with care:
- One question. Scanned once.
- "Best X" questions invite lists.
- The scan shows what the engines cited. Not why.
How to measure AI search optimization
Google: Search Console.
The Generative AI performance report shows impressions from AI Overviews and AI Mode. By page, country, device and date. With an export button.
All websites have it since 31 August 2026.
It reports impressions. Google says it may add metrics. And you may not see it if your site has too few impressions in these features.
How to read it: track Google AI Overviews.
Other engines: run your own prompts.
- Write 10 prompts.
- Run the same wording on each engine. On a fixed day each week.
- Log four things: does the answer name you? Cite you? Which URLs does it cite? Who else does it name?
AEO metrics explains the difference between a mention and a citation.
Analytics: a floor, not a total.
Our own site, 1 July to 30 September 2026 (DataFast): 29 visitors from ChatGPT. 5 from Claude. 3 from Perplexity. Out of 5,375.
A visitor with no referrer lands in Direct. So these counts are a floor.
A tool.
The AI SEO Tracker report. $49, once.
It writes up to 50 prompts from your site. You confirm them. It scans ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews. Daily for 7 days. One brand.
You get mentions, citations, position in the answer, competitor gaps, the sources each engine cited and a prioritized action plan.
It doesn't scan Google AI Mode, Claude, Grok, DeepSeek or Meta AI. Use Search Console for AI Mode.
One warning. Google says no third-party tool has access to its internal ranking or AI systems. That includes ours. A tool records the answers.
How to judge AI search optimization services
Google's page on hiring an SEO lists optimizing for generative AI among the services SEOs offer.
Its hiring advice applies to anyone who sells AEO or GEO.
Ask six questions:
- Can you show examples of your previous work and success stories?
- What results do you expect, in what timeframe, and how will you measure them?
- Is your AEO or GEO advice aligned with Google's official guidance, and do you cite it?
- Do you guarantee a ranking? Google says no one can guarantee a #1 ranking, so a yes is a reason to walk away.
- Will you share every change you make and the reasoning?
- Can I start with read-only Search Console access? Google advises read access only at the audit stage.
Add one of your own: ask to see the raw answers behind any visibility score.
An AI search optimization checklist for this week
The strategy in one line: be indexed, answer the sub-questions, say something only you can say, and get named where the engines read.
- Write 10 prompts a buyer would type. Be specific. "Best CRM for solo founders" works. "CRM software" doesn't.
- Check eligibility. For each page that should answer: it's indexed. The Search generative AI control is set to include. robots.txt allows OAI-SearchBot. The answer sits in the HTML text. The free Page Inspector extracts the title, headings and visible text of a public URL, so you can see what AI can read.
- Fix your top five pages. Put the answer in the first paragraph. Add what only you have. Replace vague phrases with a name or a number you can source. Cover the follow-up questions.
- Measure. Open the Generative AI performance report. Note your baseline. Run your 10 prompts on each engine, or run them with AI SEO Tracker. Run the same prompts again on a fixed date.
How soon does a change show? No source gives a reliable timeline.
The full first-week list: AI search engine optimization.
