Getting cited in AI answers: what Google's own guide says to do.

Somebody in the business types a question into Google, reads the AI Overview at the top and notices that three competitors are quoted and your website is not. Or they ask ChatGPT for a shortlist of suppliers in your category and the answer names everyone but you. The question that follows is a fair one: what do we have to do to be cited? Until recently the answers came from vendors and consultants. In May 2026 Google published its own guide to succeeding in generative AI features, and it is more direct than most people expected.

This article summarises what the guide says, adds what OpenAI documents for ChatGPT search, and translates both into work a marketing team can schedule.

How AI answers choose their sources.

Google describes its generative AI features as being rooted in the same ranking and quality systems that run ordinary Search. Two techniques sit on top. The first is retrieval-augmented generation, which Google also calls grounding: the model retrieves relevant, up-to-date pages from the Search index, reads specific information from them, and shows clickable links to the pages that support the response. The second is query fan-out, which Google defines as a set of concurrent, related queries generated by the model to fetch additional results for the user's question.

Google's own example is useful. A person asks how to fix a lawn full of weeds. The system may also search for the best herbicides for lawns, how to remove weeds without chemicals and how to prevent weeds. A page that answers one of those narrower questions well can be cited in a response to the broad one, even though nobody typed its exact topic.

Two consequences follow. A page does not need to rank first for the headline query to be cited; it needs to be the clearest source for one of the questions underneath it. And a page that only restates what every other result says gives the model no reason to choose it.

The eligibility rules are the ones you already know.

Google states that there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations. To be eligible as a supporting link a page must be indexed, eligible to appear in Google Search with a snippet, and meet the standard technical requirements. Google also notes that the website must be included in generative AI features in Search Console, which is the default setting.

That means the ordinary hygiene items decide eligibility before any content quality question arises: crawling allowed in robots.txt and by the hosting or CDN layer, important content available as text, internal links that make pages findable, and a page experience that does not frustrate the reader. Google adds that content rendered by JavaScript is processed as long as it is not blocked. Google also repeats its standing caveat that meeting every requirement does not guarantee a page will be crawled, indexed or served.

Non-commodity content is the whole game.

The most important line in the guide is also the least technical. Google says that creating content people find unique, compelling and useful will influence presence in generative AI search over the long run more than anything else it recommends. It contrasts commodity content, its example being a generic list of tips for first-time home buyers, with non-commodity content that offers a unique expert or first-hand perspective beyond common knowledge.

For an agency, a manufacturer or a professional services firm this is good news, because you hold material that no model can generate: the decision you made on a real project and why, the number that surprised you, the mistake that cost a client a month, the way a rule applies in Victoria rather than in general. Google explicitly warns against recycling what others have already said or what a generative AI model could produce on its own.

Practically, a page earns a citation when it does at least one of the following.

• States a clear answer near the top, in plain language, before the background. The In brief block on this article is our version of that habit.
• Adds first-hand evidence: a figure from your own data, an observation from a real engagement, a photograph you took.
• Takes a position. A comparison that concludes is more useful to a model and a reader than one that lists pros and cons and stops.
• Is organised for a reader. Google asks for paragraphs, sections and headings that give the page a clear structure, and for images and video where they help. Headings that read as the questions people ask make each section quotable on its own.

What Google says to stop worrying about.

The guide is unusually blunt about tactics. It says structured data is not required for generative AI search and there is no special schema.org markup to add, although structured data remains worth keeping for rich results. Its closing recap tells website owners that, for Google Search, they can ignore tactics such as chunking content, creating unnecessary AI text files, naming llms.txt as the example, and pursuing inauthentic mentions.

It also draws a policy line. Producing a separate page for every variation of how people might phrase a question, including the fan-out queries themselves, primarily to influence rankings or AI responses falls under Google's scaled content abuse spam policy. A high quantity of pages does not make a website more relevant. Google says its systems now understand relevance even when the query and the page do not match word for word, so the old habit of one thin page per keyword has stopped paying and started costing.

There is a caution for buyers of services, too. Google says no third-party tool has access to its internal ranking or AI systems, and asks businesses to evaluate any answer engine or generative engine optimisation advice against its official guidance. Google's own summary is that optimising for generative AI search is optimising for the search experience, and is therefore still SEO.

Our position on llms.txt is pragmatic. We maintain one on our own website because some assistants and agents read it and it costs nothing to keep accurate, but we do not sell it as a way to be cited, and Google has now said plainly that it does not need it.

ChatGPT search follows different rules.

Google's guide covers Google. ChatGPT search is a separate index with its own crawler, and OpenAI documents the controls. OAI-SearchBot is the crawler used to surface websites in ChatGPT search results. Websites that opt out of it are not shown in ChatGPT search answers, although they can still appear as navigational links. OpenAI recommends allowing OAI-SearchBot in robots.txt and allowing requests from its published IP ranges, and says a robots.txt change can take around 24 hours to take effect in search results.

The training crawler, GPTBot, is a separate setting. A business can disallow GPTBot to keep its content out of model training and still allow OAI-SearchBot so it appears in ChatGPT answers. We covered the crawler decisions in detail in our article on whether to block AI crawlers, and the same logic applies here: decide per crawler, write the decision down, and check the server logs afterwards.

Once access is settled, the content principles converge. A page with a clear answer, first-hand detail and a sensible structure is the page most likely to be quoted, whichever system is reading it.

Measure it in Search Console, not in screenshots.

Google now reports generative AI performance in Search Console. The guide points website owners to the Generative AI performance report to see how people discover content through AI features on Google Search and Discover. That report, rather than a folder of screenshots of AI Overviews, is where a monthly review should start. Our earlier article on measuring visibility in AI search explains how to read it.

For ChatGPT and other assistants, the evidence is in referral traffic and server logs: sessions arriving from chatgpt.com, and requests from OAI-SearchBot and the user-initiated fetchers. Neither is a ranking report, but both tell you whether the door is open and whether anyone is walking through it.

A working plan for the next quarter.

Most businesses do not need a new program. They need to point an existing one at the right pages.

• Confirm eligibility once: robots.txt, CDN rules, Search Console settings, and a crawl of the service and category pages that matter commercially.
• Pick the ten pages that carry the most revenue and rewrite each opening so the answer comes first, in one or two sentences, followed by the evidence.
• Add one piece of first-hand material to each of those pages: a real example, a measured result you are permitted to share, or a decision rule you actually use.
• Retire or consolidate the thin variants. If three pages answer the same question with different phrasing, keep the best one and redirect the others.
• Review the Generative AI performance report and your referral sources monthly, and change one thing at a time so you can see what moved.

None of this is exotic. That is the point of Google's guide. The businesses that will be cited are the ones with something worth citing, presented so that a person and a model can both find it quickly.

How we can help.

Emote Digital builds and supports websites for established Australian businesses and runs the search and content programs behind them. If you want an honest read on why competitors are being cited and you are not, we will review eligibility, the commercially important pages and the Search Console reporting, then agree a short list of changes with you before anything is written.