How to Show Up When Customers Ask AI Instead of Google (2026)

Google's own documentation is unusually blunt about this: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add." That sentence sits in Google's guidance on AI features and your website, and it quietly rules out most of what is currently being sold as answer engine optimisation. The same page states the eligibility rule just as plainly: to appear as a supporting link in AI Overviews or AI Mode, "a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements."

So the honest headline is that there is no separate AI channel to optimise for on Google's side. There is your site, and whether it is indexable, readable and worth quoting.

That said, "there's no magic file" is not the same as "there's nothing to do". Assistants outside Google run their own crawlers, honour their own robots.txt tokens, and make their own decisions about what to cite. Plenty of Australian business sites are currently invisible to at least one of them for a reason that takes ten minutes to find and about the same to fix.

Why it suddenly matters

The behaviour shift is real, and it is measurable without anyone having to invent a statistic. Pew Research tracked the browsing of 900 US adults across 68,879 Google searches in March 2025 and found that when an AI summary appeared, users clicked a traditional search result in 8% of visits, against 15% when no summary was present. Around one in five searches in that dataset produced an AI summary at all.

Roughly half the clicks, on the queries where the answer box shows up. We've written separately about what that does to a traffic graph while rankings stay perfectly still. The relevant point here is the flip side: if the answer is doing the work the click used to do, being inside the answer is worth something, and being absent from it costs something.

The three different things people mean by "showing up in AI"

They get bundled together, and they have completely different levers.

The first is Google's AI Overviews and AI Mode. These draw on Google's existing index. Eligibility is ordinary Search eligibility — indexed, snippet-eligible, technically sound. There is no opt-in, and the only controls Google offers point the other way: nosnippet, data-nosnippet, max-snippet and noindex all remove you. Nobody can sell you inclusion, because there is no inclusion mechanism to sell.

The second is assistant search — ChatGPT's search feature, Perplexity, Claude's search. These run their own crawlers against their own indexes, and each publishes a robots.txt token that governs whether you can appear. This is the layer where a site can be genuinely, silently locked out.

The third is a live fetch: someone asks an assistant about your business by name, or pastes your URL into it. The assistant goes and reads the page in that moment. What matters then is not indexing at all — it is whether the page states, in plain text a machine can parse, what you do, where you do it, what it costs, and how to start.

Who is actually allowed to read your site

This is the check worth doing first, because it is binary and it is cheap. A lot of sites blocked AI crawlers wholesale during 2024 and 2025, either deliberately or because a plugin or host shipped a default, and never revisited the decision once "training crawler" and "search crawler" became different things.

The practical version: open yourdomain.com/robots.txt and read it. If you find a blanket Disallow: / against OAI-SearchBot or PerplexityBot, you have chosen to be absent from those products, and you should at least have chosen it on purpose. Sources: OpenAI's crawler documentation, Perplexity's bot guide, and Anthropic's crawler policy.

Entity clarity, which is a fancy way of saying "say what you are"

Once a machine can read you, the question becomes whether it can work out what you are with enough confidence to name you in an answer. This is where the useful, unglamorous work lives, and it looks almost nothing like SEO tactics.

A model assembling an answer to "good physiotherapist in Geelong who does dry needling" is trying to match an entity to a set of attributes. It needs your business to be one consistent thing across every place it appears — same legal and trading name, same address in the same format, same phone number, same description of the service. Where those drift, confidence drops, and the assistant reaches for a directory listing or an aggregator instead, because the aggregator is internally consistent even when it is out of date.

Then there is the writing itself. Most Australian business sites describe what they do in language that survives a board meeting and communicates nothing: solutions, tailored approach, passionate about outcomes. None of that is quotable. A page that says "we're a physiotherapy clinic in Geelong West; consultations are 40 minutes; dry needling and shockwave therapy are available; initial appointment is $110" contains five extractable facts. The vague page contains none. Assistants quote sentences, so write sentences worth quoting — specific, checkable, and about your actual business rather than your category.

Third, put the facts on your own site rather than only on Instagram or a booking platform. If your opening hours, prices and service area exist only inside a third-party widget or a social profile, the assistant will cite the third party. That is a straightforward way to hand your own name to a platform that then charges you for the lead. Some of this overlaps with why a site fails to surface in ordinary search too — we've covered that ground in why your website isn't showing up on Google in Australia.

llms.txt: worth publishing, not worth paying a retainer for

We publish one at prycedigital.com/llms.txt, so this is a recommendation we follow. It is also one we're not going to oversell.

llms.txt is a convention proposed by Jeremy Howard in September 2024: a markdown file at the root of your site with a heading, a short summary of what the organisation is, and curated links to the pages that matter. The reasoning is sound — a model landing cold on a site has to infer structure from navigation and guesswork, and a clean summary saves it the trouble.

The honest state of play is that publishing it is well ahead of anything consuming it. The spec itself uses the future tense: agents "are expected to" read the file. Chrome's Lighthouse now includes an llms.txt audit, but marks a missing file as Not Applicable "as providing the file is optional at the moment," and describes the whole thing as an emerging convention. And Google, as quoted at the top of this piece, says explicitly that you don't need to create AI text files to appear in its AI features.

So: write one, because it takes an afternoon, it forces you to articulate what your business is in a hundred words, and it costs nothing if the convention takes hold. Do not pay anyone a monthly fee to maintain a static text file, and do not let it be presented to you as the reason your enquiries will improve.

What is snake oil

Guaranteed placement in AI Overviews. There is no inclusion mechanism, only exclusion controls, so nobody can guarantee it and anyone claiming otherwise is describing a product Google does not offer.

"AEO audits" that turn out to be a standard technical SEO crawl with the headings renamed. If the deliverable is broken links, missing meta descriptions and image alt text, that is an SEO audit, and it is worth having under its own name at its own price.

Prompt-stuffing — writing pages full of "questions people ask ChatGPT" in the hope of matching a query string. Assistants are not doing keyword matching, and a page built for an imagined prompt reads badly to the humans who eventually land on it.

Paid inclusion in "AI citation networks". Whatever is being sold there is a link scheme with new packaging.

And schema markup positioned as the AI lever. Structured data is genuinely useful — it removes ambiguity about your address, hours, prices and reviews, and it drives rich results in ordinary Search. It is just not the thing standing between you and an AI answer, and Google has said so directly. Do it properly, for the right reason.

Where to start

The order that actually helps: read your robots.txt and decide, deliberately, which crawlers you want reading you. Confirm the pages you care about are indexed and snippet-eligible in Search Console, because on Google's side that is the eligibility test. Make the page fast and cleanly structured, so both crawlers and people get through it. Then rewrite the copy so the facts about your business — what, where, how much, how long, who it's for — are stated in plain sentences on pages you own, not implied across a gallery and a booking widget.

Almost all of that is the same work that makes a site perform in normal search, which is the least exciting and most reliable conclusion available. It is also the kind of thing that gets built in from the start on a properly built site rather than retrofitted after a platform has already made the decisions for you.

If you want a fast read on the technical half of that list, run the free audit. It checks Core Web Vitals, your certificate, DNS and email records, HTML structure, page weight and crawl readiness including robots.txt, in about thirty seconds. It won't tell you whether ChatGPT likes you — nothing honestly can — but it will tell you whether the machines can read the page at all, which is where every version of this conversation has to start.