SEO for AI search: what Google actually asks of you
Google's answer to the "AI search" sales pitch is short: the AI answers in Search are built on the same index and ranking systems as the ten blue links. Here's how AI Overviews and AI Mode pick the pages they link to, and what you can safely ignore.
By the Boltly team · about 13 minutes to read
Updated
SEO for AI search is still SEO
SEO for AI search means making your pages eligible and worth linking to when Google writes an AI answer. Google's guide to its generative AI features says the best practices for SEO still apply, because those features are rooted in its core Search ranking and quality systems. A page has to be indexed and allowed to show a snippet. After that, the same things that earn a normal ranking decide whether it gets picked.
"AEO" (answer engine optimisation) and "GEO" (generative engine optimisation) are, in its words, still SEO when the target is Google Search. If someone sells you either as a separate discipline, Google points you to its page on judging third-party SEO advice.
Two things have changed: which questions earn a click, and a new Search Console report that counts how often your pages appear inside AI answers.
Sources: Google Search Central, generative AI features guide, last updated 10 July 2026; "AI features and your website", last updated 10 December 2025. Both read 4 October 2026.
Retrieve first, write second
Google names two techniques behind its AI answers, query fan-out and grounding, and both start from the normal index.
- Step 1
Your page is crawled and indexed
Googlebot fetches the page and stores it, the same as for any search. If robots.txt, a CDN rule or a firewall blocks Googlebot, the page can't be used anywhere in Search, AI answers included.
- Step 2
The question fans out
For a complex question the model issues several related searches at once. Google calls this query fan-out. Each related search is answered by the core ranking systems, so ranking for the narrower questions matters.
- Step 3
Pages are retrieved
Google describes this as retrieval-augmented generation, also called grounding. The ranking systems fetch relevant, current pages from the index, and the model reads the specific information on them.
- Step 4
The answer is written, with links
The response is generated from what was retrieved and shows links to pages that support it. AI Overviews and AI Mode may use different models, so the links they show for the same topic can differ.
Query fan-outHow one question becomes several searches
- hot water system lifespan
- repair or replace hot water
- heat pump vs gas hot water
- hot water rebates in my state
An illustration of the technique Google describes in its AI features guide. The sub-searches are our example, not real query data.
What this means in practice
Because the fan-out searches are ordinary searches, a page that answers one narrow part of a bigger question well can be linked even if it would never rank for the big question itself. A plumber's page on how long a gas storage hot water system lasts, and the warning signs of failure, is the sort of page pulled into an answer about repairing or replacing one.
It also means there's no back door. If a page doesn't rank anywhere in normal search, nothing in Google's documentation suggests the AI layer will find it.
Not every search gets an AI answer
Google says AI Overviews appear only when its systems judge them additive to classic Search, and often don't trigger. Simple navigational searches, like a business name, usually show normal results.
AI Mode is a separate tab built for longer questions that need comparison or reasoning. Google says it rolled out to Australians from 8 October 2025.
Sources: Google Search Central, "AI features and your website"; Think with Google APAC, "Google AI Mode for marketers in Australia", October 2025.
The seven checks Google lists
Google's "AI features and your website" page says no special work is required, then lists seven SEO basics. Here they are as checks you can run this week.
- Crawling is allowed. Check robots.txt, and ask your host whether a CDN or bot-protection setting is challenging Googlebot. Security tools sometimes block crawlers silently.
- Pages are reachable through internal links. A page linked only from the sitemap is hard to find. Link to it from the pages a visitor would read first.
- The page is pleasant to use. It loads quickly on a phone, the main content is easy to tell apart from ads and pop-ups, and it works on every screen size.
- Important content is text. Prices, opening hours and service areas inside an image or a PDF are harder to use than the same facts written on the page.
- Images and video support the text. AI answers can show them. Good alt text and a descriptive file name help.
- Structured data matches what's visible. Markup that says one thing while the page says another is a problem in ordinary search too.
- Business Profile and product feeds are current. Google says AI answers can draw local details from Business Profile and product information from its merchant feed service.
None of these are new, which is Google's point. If your site passes them, the technical part is done; our technical SEO guide walks through any it fails.
The part that matters most: non-commodity content
Google's generative AI guide says that creating unique, useful content will likely influence your presence in AI search more than any other suggestion it makes. It draws a line between commodity content, built from common knowledge that anyone could write, and non-commodity content that adds first-hand experience or expertise.
"7 signs your hot water system needs replacing"
Common knowledge rearranged. A model can write this from what's already online, so there's little reason to link to yours.
"What we found inside 40 failed hot water tanks"
Only a business that did the work can write it: photos, the failure patterns it saw, what it would do differently. This is what an AI answer needs a source for.
The two titles are our illustration of Google's distinction, not real articles. Google's own example compares "7 Tips for First-Time Homebuyers" with a first-hand account of a sewer line inspection.
Six things Google says you can ignore
Google's generative AI guide has a section on what you don't need to do. If a quote or audit lists any of these as AI search work, ask why.
An llms.txt file
Google Search doesn't use llms.txt or any other "AI text file". Creating one won't help or harm your rankings. It may matter to other services, which is a separate decision.
Special AI schema
Structured data isn't required for AI features and there's no special schema.org type for them. It's still worth keeping for rich results.
Chunking your pages
There's no need to cut content into tiny pieces so a model can read it. Google says its systems understand pages covering several topics and show the relevant part.
Rewriting copy for robots
You don't need a special writing style for AI answers, or every long-tail variation of a phrase. The systems understand synonyms and meaning.
Buying "mentions"
Seeking inauthentic mentions across the web isn't as helpful as it looks, Google says. Its spam systems filter what its AI features rely on.
Pages for every fan-out query
Making a separate page for each variation people might search, mainly to sway AI answers, breaches Google's scaled content abuse policy.
Source: Google Search Central, generative AI features guide, sections on mythbusting and focusing on what users want, last updated 10 July 2026, read 4 October 2026.
Search Console now counts AI impressions
Google's Search Console help says its Generative AI performance report had rolled out to all websites worldwide by 31 August 2026.
The report counts impressions: how many times a link to your site was shown in AI Overviews or AI Mode. You can group the numbers by page, country, date or device, and filter between typed searches and image searches from tools like Google Lens. It doesn't show clicks or the questions people asked, so treat it as a visibility gauge, not a traffic report.
Clicks from AI features still sit inside the normal Performance report under the Web search type. Pair both with Google Analytics to see what visitors did next. Google says clicks from results pages with AI Overviews tend to be higher quality, meaning people spend more time on the site. Your analytics will show whether that holds for you.
Generative AI performanceSearch Console
Measures: impressions in AI Overviews and AI Mode
- Pages
- Countries
- Dates
- Devices
- Search typeWeb: text-based, or Web: multimodal
- Not shownClicks, queries, positions
- WherePerformance section of Search Console
- Related settingSettings, then Search generative AI
Outline only. Bars are decorative and show no data.
Share of visits where people clicked a traditional result, with an AI summary on the page versus without one.
Pew Research, July 2025 (US browsing data)Share of US Google searches from January to April 2026 that ended without a click to any website.
SparkToro with Similarweb, 8 June 2026Date by which Google says the Generative AI performance report reached every Search Console property.
Search Console Help, read 4 October 2026The two click studies are American, and Australian figures haven't been published to the same standard. They point the same way, though: simple questions get answered on the results page, so the clicks that remain go to pages offering something the summary can't.
Opting out, and what it costs you
Google gives you two ways to limit AI use of your content, and both cost you something.
The broad option is the Search generative AI control in Search Console, under Settings. Google's help page says it has been available to every site since 31 August 2026. The default is to include your site. If you switch to exclude, your links and content stop appearing in AI Overviews, AI Mode and Discover's generative features, usually within a few days. Google states the setting isn't used as a ranking signal for the rest of Search, and it doesn't stop AI training, which is handled separately through Google-Extended.
The narrow option is page level. The nosnippet, data-nosnippet and max-snippet controls limit what Google can show from a page, in AI features and in normal snippets alike. A noindex removes the page from Search entirely.
For most small businesses we'd leave the default on. Excluding your site doesn't stop the AI answer appearing; it's built from someone else's pages instead, which Google's help page warns may look similar to yours.
For local businesses, the Business Profile does a lot of the work
When someone in Ballarat asks Google for a mechanic open on Saturday, the answer leans on business details: hours, categories, location and reviews. Google's AI guide names Business Profile, along with product feeds for retailers, as a way to be visible in AI responses as well as normal results. Keeping your profile complete and correct is the most direct AI search task a local business can do. Our Google Business Profile guide covers the setup, and the local SEO explainer shows how it fits with your website.
A sensible order of work
- Confirm Googlebot can crawl the site and that key pages are indexed (Search Console's Page indexing report).
- Fix your Business Profile or product feed so the facts match your website.
- Rewrite your thinnest service pages with things only you know: prices you actually charge, jobs you've done, mistakes you see customers make.
- Open the Generative AI performance report monthly and note which pages appear. Compare with Analytics.
- Ignore anyone selling llms.txt files, AI schema or bulk "mention" packages as Google work.
Source: Search Console Help, "Search generative AI control" and "Generative AI performance report (Search)", read 4 October 2026.
Fewer clicks, a new "GEO" invoice, and visits from chatbots you can't see
Three questions come up most once AI answers start showing for a business's searches, and each is answered by looking at the right number.
Impressions are up, clicks are down, and rankings look the same
Open the Performance report in Search Console, compare the last three months with the same months a year earlier, and switch on clicks, impressions, CTR and position. If position held steady and impressions rose while CTR fell, the searchers saw you and got their answer on the page. That's the pattern the Pew and SparkToro studies above describe, not a penalty. Confirm the Manual actions report is clear, then filter by query: if the drop sits on "how" and "what is" searches rather than on the searches that bring enquiries, judge the month by calls and forms, not sessions.
Your agency wants an extra monthly fee for "GEO" or "AI visibility"
Run the line items through the checker at the top of this page. Anything on Google's skip list isn't Google work. Anything that is ordinary SEO, such as crawl fixes, a better Business Profile or pages with first-hand detail, should already be in the SEO you pay for. Then ask how they'll measure it. If the answer is a score from their own tool, ask how often its answers change between runs.
You want to know whether ChatGPT or Perplexity sends you anyone
In Google Analytics, open the traffic acquisition report and look at session source. Visits from people who click a link in an assistant can show up as referrals from domains such as chatgpt.com or perplexity.ai, though some arrive with no referrer and land in direct. Small-business numbers are usually small and patchy, so add a "how did you find us?" question to your enquiry form.
SEO for AI search, answered
What is SEO for AI search?
It's the work of making your pages eligible and worth linking to when Google writes an AI answer. Google says its AI Overviews and AI Mode are rooted in its core ranking systems, so a page must be indexed and snippet-eligible, then compete on the same quality signals as any search result.
How do I improve content for AI Overviews?
Google says there are no extra requirements. Let Googlebot crawl the page, link to it internally, keep important facts in text, and make the content more useful than what's already out there. First-hand detail (real prices, jobs and photos) gives an AI answer a reason to link to you.
Does llms.txt work for Google?
No. Google's generative AI guide says Google Search doesn't use llms.txt or similar AI text files, so adding one neither helps nor harms your visibility in Google. Other services might read it; that's a separate choice.
Is SEO dead because of AI?
Google's own documentation says the opposite: the best practices for SEO still apply to its AI features because those features retrieve pages through the core ranking systems. What has changed is which searches end in a click, so pages that only restate common knowledge lose out.
How do you rank in Google AI Overviews?
There's no separate ranking to chase. AI Overviews link to pages that the ranking systems retrieve for the question and its related fan-out searches. Ranking for narrower questions in your field gives you more chances to be linked. Nobody can promise a link in an AI answer.
Do I need schema markup for AI Overviews?
No. Google says structured data isn't required for its generative AI features and there's no special schema type for them. It can still earn some rich results if the markup matches the visible page. Our schema markup explainer covers which types are worth adding.
Is Google AI Mode available in Australia?
Yes. Google's Think with Google APAC site says AI Mode began rolling out to Australians from 8 October 2025.
Google's autocomplete suggested "seo for ai search", "does llms.txt work", "is seo dead because of ai" and "how to rank in google ai overviews" when we searched from Australia on 4 October 2026.
Want a second opinion on your AI visibility?
Our free audit checks crawling, indexing and the pages most likely to be linked, against Google's own guidance.