Getting found by AI search

Two ways your site gets seen now

For most of the web's life there was one prize: a link on page one. Someone searched, your link showed up, they clicked, you got a visit.

That prize still exists. But there is a second one now, and it works differently. A growing share of searches end with an answer instead of a list: Google paints an AI Overview above the blue links, ChatGPT returns a sourced paragraph, Perplexity stitches three cited sites into a summary. The person gets what they wanted without clicking anything.

So you are optimising for two surfaces at once. The classic one (blue links) and the answer one (AI-generated responses). The good news, which most of the panic online ignores, is that they share almost all the same plumbing. This guide is the map: what each surface actually rewards, where Google differs from everyone else, and the concrete moves that put you in both.

For Google, AI search is just SEO

Here is the single most useful thing to understand, because it cuts through a lot of noise.

Google's AI features do not run on a separate index or a secret AI-only crawler. AI Overviews and AI Mode read from the same core Search index everything else does. Under the hood they use Retrieval-Augmented Generation (the model retrieves real indexed pages and writes its answer from them) and a Query Fan-Out technique that fires off a cluster of related searches at once and pulls candidate sources from all of them.

The practical consequence: if a page is indexable, helpful, and well-structured, it is already eligible to be one of those sources. Google's own AI optimization guidance says it plainly. There is no "AEO" or "GEO" trick for Google's AI; it is the same SEO you should already be doing.

Which also means the things people invent for AI do nothing for Google here. You do not need an llms.txt file, custom AI markup, or content chopped into tiny "AI-friendly" chunks for Google. You should not rewrite pages specifically for the model, chase inauthentic mentions, or spin up a separate page for every query variation (Google calls that scaled content abuse). We pull this apart in AEO vs SEO: what changes when AI answers, and the source-selection side specifically in how AI Overviews pick their sources.

If you want the foundation that makes all of this work, start with SEO for founders. It is the same three steps Search has always run on: crawl, then index, then serve.

The honest nuance: Google is not the whole answer ecosystem

Everything above is true for Google. It is not the full story, because Google is not the only answer engine your customers use.

ChatGPT, Perplexity, Claude, and the wave of AI agents that browse on a user's behalf do not all share Google's index or Google's rules. Some have their own crawlers. Some lean on different retrieval pipelines. And in that wider world, emerging conventions and raw content legibility carry more weight than they do with Google.

This is the one place the advice forks:

• For Google's AI: indexable, helpful, well-structured. No llms.txt, no special markup.
• For the wider ecosystem: the same fundamentals, plus making your content easy for any model to read and quote, plus opt-in conventions some crawlers now respect.

llms.txt is the clearest example of that fork. It is a plain-text file that points AI crawlers at your important pages. It does nothing for Google, full stop. Whether it earns its keep anywhere else is genuinely unsettled, which is why we wrote an honest assessment in does llms.txt do anything rather than a sales pitch. The rule we hold to: never claim llms.txt helps Google, and never overpromise what it does elsewhere.

What actually puts you in the answers

Strip away the jargon and the work is concrete. Most of it is plain quality.

Be crawlable and indexable. A page that Googlebot cannot reach or render cannot be retrieved, full stop. Google is mostly mobile-first now, so the primary crawler is Googlebot Smartphone, and it reads roughly the first 2MB of an HTML page. Make sure the crawler can fetch the same CSS and JS your visitors get, and check what is actually indexed with a site:yourdomain.com query in Google.

Make the text real text. This is the one that silently kills AI visibility. A price baked into a screenshot, a key spec trapped in an image, an answer that only exists inside a PDF: a model cannot quote what it cannot read. If you cannot select the text with your cursor in a browser, it probably is not being parsed. We go deep on this in making your content legible to AI.

Answer the actual question, clearly. AI answers get assembled from passages that directly and self-containedly address a query. Pages that lead with a clean answer, then support it, get quoted more than pages that bury the point. This is helpful-content thinking: write for the person, be clear about who wrote it and how, and earn trust. (E-E-A-T is real guidance, but note it is not a direct ranking dial you can turn.)

Use structure for clarity, not for tricks. Logical headings and semantic HTML help any reader, human or model, find the section that answers a question. But do not over-index on it: heading order does not change Google rankings, and structured data is a useful hook, not a requirement. Structured data for AI search covers where schema genuinely helps and where it is just busywork.

Earn citations the slow way. In the wider ecosystem, getting named in answers tracks closely with being a credible, frequently-referenced source. There is no shortcut, and faking mentions backfires. Getting cited by ChatGPT and Perplexity is about doing the legitimate version of that.

Where to start

If you only do one thing this week: open your most important pages and try to select the most important text with your cursor. Anything you cannot select is invisible to AI answers. Fix that first, because it is the highest-leverage, lowest-effort move there is.

Then work the fundamentals from SEO for founders outward. Get crawled, get indexed, answer real questions in real text. That single foundation carries you across both surfaces: the blue links Google has always served, and the AI answers it (and everyone else) is serving now.

Start with the fundamentals. Extend honestly for AI. Do not buy any trick that promises to skip the work.