How to Show Up in AI Answers: The Path in Order, Through the Four Gates of Citation

The owner of a clinic types into Google the question her patients ask at the front desk. Before any link, a finished paragraph appears. It answers calmly and in detail. Next to it, three sites show up as sources. None of them is hers.

She does not want to be the fourth link. She wants that paragraph to have come from her site, with her name attached.

Google itself separates those two spots, on a Search Console help page. A site’s content can appear in the AI features of Search as a link. It can also help ground the answer the AI writes. ¹ This piece walks the path to both spots, in the order the gates close. Every step has an official document behind it. Where independent measurement disagrees with the owner of the system, I show the disagreement.

How the ready-made answer gets assembled

Picture the clerk at a hardware store. You ask which grout holds up in a swimming pool. He does not know offhand. He pulls the catalogs of three manufacturers, finds the right passage, and answers in his own words. If you ask, he shows you which catalog it came from.

The AI in Search makes the same move. In Google’s description, its AI features use the core ranking systems to retrieve relevant, up-to-date pages from the index. Then they review the information on those pages and write the answer, with links to the pages that support it. ² This is called retrieval-augmented generation. Google also calls it grounding.

Two consequences follow for anyone who runs a business. A catalog that is not on the shelf never gets consulted. And the catalog that does get consulted needs the right passage, readable, with the manufacturer’s name on the page.

The four gates, told from the writer’s side

In chapter 9 of my book SEO de Entidade, published in Portuguese, I organize the path to a citation into four gates. Being in the index the system consults. Having a passage that serves the question. Surviving the reranking of candidates. Carrying a claim that can be cut out and attributed to you. ³ The organization into four is mine. The steps are well-known engineering in retrieval systems.

A separate piece on how an assistant decides whom to cite shows that line from the machine’s side, with the symptom at each stop. Here I tell it from the side of the site owner.

The third gate happens inside the machine, and no company publishes its own reranker. ³ What is left for the writer is what the page needs to be at the other three. The last one asks for two qualities at once. The count comes to four requirements, in this order: retrievable, relevant, extractable, and attributable. Polishing the fourth while the first is still closed is wasted work.

1. Retrievable: can the machine reach your text?

The first question is almost boringly basic, and it is where most people get stuck. For Google, the condition is in writing. The page has to be indexed and eligible to show in Search with a snippet, a short piece of its text. ^{2 4}

If the page is not even indexed, the problem comes before any AI. The diagnosis of the stages of search is in how to get your site on Google.

There is a second switch, rolled out to every site on August 31, 2026. In Search Console, each property has a Search generative AI control. It covers AI Overviews, AI Mode, and the generative AI features in Google Discover. ¹ The default is to include your site. A site switched to exclude drops out of the links and stops grounding answers in those features. ¹

The same help page has a sentence every business owner should read before touching this setting. With the site excluded, “Content from other sites will still be available in those features, and it may appear similar to yours.” ¹ The answer keeps existing. It just gets written with material from your neighbor.

Look closely at the dates. Google’s December 2025 page on AI features still says “There are no additional technical requirements.” ⁴ The July 2026 guide added the Search Console control as a condition. ² Official documentation ages too, which is why every note here carries an access date.

Blocking can also live outside your site. Google’s documentation says to make sure crawling is allowed in robots.txt and by any CDN or hosting infrastructure. ⁴ In practice, that is a dashboard the marketing team almost never opens.

And Google is one system among several. OpenAI writes that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. ⁵ Microsoft Copilot runs on Bing’s search index. ⁶ And Bing states that it respects the preferences site owners express in robots.txt. ⁷ Each system has its own entrance. Yours may be open in one and closed in another.

2. Relevant: do you answer the question the machine went looking for?

AI rarely searches only for the phrase the person typed. Google gives an example in the July guide, “how to fix a lawn that’s full of weeds.” That question sets off other searches: “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in lawn.” ²

The model generates those related queries at the same time, to fetch more information. ² This is called query fan-out. The piece on how an assistant decides tells the story of the term.

Each smaller question has its own winners. For your business, relevant means having a page that answers one of those smaller questions in full.

Bing shows those questions from the inside. The AI Performance report in Bing Webmaster Tools lists grounding queries, the phrases the AI used to retrieve the content it cited. ⁷ Since June 2026, it groups those phrases by intent and by topic. ⁸ That is the closest a site owner gets to seeing the question the machine rewrote.

There is a temptation here, and Google shut it down in writing. Creating a page for every variation of a question, primarily to manipulate AI answers, violates the scaled content abuse spam policy. ² The same guide adds that a high quantity of pages does not make a site more relevant. ²

The path the measurements point to is coverage. Kevin Indig analyzed more than 21,000 citations to measure the effect of content length, depth, and focus. ⁹ The full study sits behind a paywall. The finding my book records from it: breadth of topic coverage weighed more than domain authority in the choice of sources. ⁹

Covering your territory means answering, with pages that deserve to exist, the questions that surround what you sell. A hundred near-identical pages sit on the other side of the line Google drew.

3. Extractable: can your passage survive leaving the page?

Take a pair of scissors and cut a paragraph out of your site. Tape it to a note and hand it to a stranger. Does that person understand it without reading the rest?

If the paragraph opens with “this happens because,” it depends on the one before. On its own, it answers nothing. A passage that can be cut out without breaking stands on its own. This is called an extractable passage. ³

Microsoft says this in almost the same words. In an October 2025 guide, it recommends “Sentences that make sense even when pulled out of context.” It also says long walls of text blur ideas together and make it harder to split content into usable chunks. ⁶

Here Google says something else, and the disagreement needs to stay in plain view. The July guide says “There’s no requirement to break your content into tiny pieces for AI to better understand it.” It says Google’s systems understand several topics on one page and show the relevant piece. ² And it says “You don’t need to write in a specific way just for generative AI search.” ²

The two companies agree on one point. A page organized into paragraphs and sections, with clear headings, helps the reader. ^{2 6} My reading of the two together is this. A paragraph that stands on its own serves the person first, and it does not call for a page chopped into pieces. Google speaks for Google. Microsoft speaks for Copilot.

Extractable also has a technical side. Google asks for important content to be available in textual form. ⁴ Microsoft says not to hide answers in tabs or expandable menus, because “AI systems may not render hidden content.” ⁶ A nosnippet rule on the page keeps the content from being a direct input for AI Overviews and AI Mode. ¹⁰ The max-snippet rule limits how much of the text can go in. ¹⁰ One of these rules, forgotten in the site’s theme, closes the gate without warning.

There is also a neighboring reader. The same Google guide mentions browser agents that interpret, among other things, the accessibility tree of a page. ² That reader has its own checklist.

4. Attributable: does the machine know who is talking?

This is the gate of silent frustration. Your text was read and helped build the answer. When the credit was handed out, another name showed up. ³

Compare two sentences about the same procedure, in an example built for this piece. “Most patients recover quickly” sounds friendly and backs up nothing. “In 2025, 290 of the clinic’s 340 patients were back at work within two days” travels with a return address. It carries inside it where the information came from. This is called an attributable passage. ³

Microsoft points the same way: “Examples, data, and cited sources help build trust when content is reused in AI-generated answers.” ⁷ Google asks for content that goes beyond common knowledge, with first-hand experience. ² A list of tips any site could write has no owner. The account of someone who did the work has one.

Attribution also requires knowing who you are. Microsoft recommends aligning text, images, and video so they “consistently represent the same entities, products, or concepts.” ⁷ If your company shows up with three spellings and two descriptions, the machine has reason to hesitate over the credit. That reading comes from the book. ³ The company’s official address on the web, with name, description, and relationships, is the subject of the entity home piece.

Google takes one shortcut apart in writing, and that shortcut is going out to buy mentions. The guide says “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.” ²

What Google says, and where measurement disagrees

The strongest evidence on this subject comes from Google itself. In the July guide, it names AEO and GEO and closes the passage like this: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” ²

A limit comes attached. This is Google talking about Google, and Google has a stake in the matter. The sentence says nothing about ChatGPT, Perplexity, or Copilot. Each one has its own crawler, index, and rules. ^{5 6}

The first disagreement is about position. Google writes that its AI features are rooted in its core ranking systems. ² Kevin Indig analyzed more than 546,000 AI Overviews in 2024. More than half of the citations came from outside the top 20 results. ¹¹ Query fan-out explains part of the gap. The AI went looking for other questions, and each one has its own ranking. That bridge is my reading. ³ The full dispute between position and citation, with measurements that clash, is in the piece on LLMO, GEO, and AEO.

The second disagreement is about writing. Google says nobody needs to write in a specific way for AI. ² The paper that coined the acronym GEO measured something else, on an engine the authors built themselves. Citing sources, adding quotations, and including statistics raised visibility in their test. ¹² Look at that list calmly. These are habits of good writing for people, and Google forbids none of them.

How to tell whether it is working

There is no single scoreboard. There are two free reports, each with a caveat from its owner.

In Search Console, the Generative AI performance report counts how often links to your site appeared in AI Overviews and AI Mode. It breaks that down by page, country, device, and date. ¹³ If the report does not show up, the help page lists likely reasons. The site may not have enough impressions in these features, or it may have been excluded in the control. ¹³

In Bing Webmaster Tools, the AI Performance report counts citations in Copilot and in AI-generated summaries in Bing. ⁷ In June 2026, it gained Citation Share, your site’s slice of all citations for the same grounding query. ⁸ Microsoft warns that the metric is observational. It does not work as a ranking or as a scoreboard against competitors. ⁸

Be wary of anyone offering Google’s internal numbers. The guide states it plainly: “No third-party tool has access to our internal ranking or AI systems.” ²

The order that holds from today on

When someone says your company needs to show up in AI answers, ask four questions, in this order. Can the machine reach the text? Does the text answer one of the smaller questions? Can the passage survive leaving the page? Does the machine know who is talking?

Nobody guarantees the citation. Google writes that “Indexing and serving aren’t guaranteed,” even with everything in order. ² What is in your hands is not getting stopped, through carelessness, at the gates that depend on you.

The first step is physical and takes five minutes. Open Search Console, go to Settings, then Search generative AI, and check that your site is set to include. ¹ Then write down the ten questions customers ask before they buy. Look on your site for the paragraph that answers each one, on its own.