How to get cited in Google AI Overviews, and in ChatGPT, Claude, and Perplexity

Google's own documentation, as of October 30, 2026, says there is no special trick: to be cited as a link in an AI Overview, a page has to be indexed and eligible to show in Search with a snippet, and Google says there are "no additional technical requirements," no special schema, and no new AI text files to add. What the citation is worth is the part people skip. Seer Interactive's 2026 update (published April 2026; 53 brands, January 2025 to February 2026) found that on informational queries with an AI Overview, a tracked brand's organic click-through averaged 2.07% across 2025 when its domain was cited and 0.94% when it wasn't, against 3.35% with no Overview.

So being cited roughly doubles your clicks compared with sitting under the Overview uncited. Seer notes one account holding 47% of the uncited impressions drags that 0.94% down; without it the uncited figure is about 1.61%, so the gap is closer to 1.3x. Both numbers are small, of course. A citation is the better seat in a smaller room.

The room did get smaller. Pew Research Center tracked the browsing of 900 US adults in March 2025 and found that people who saw an AI summary clicked a regular search link on 8% of visits, against 15% for people who didn't see one. That study is from 2025 and may be dated now. Ahrefs compared December 2023 with December 2025 across 300,000 keywords and found that an AI Overview "now correlates with a 58% lower average clickthrough rate for the top-ranking page." The Ahrefs result is desktop data and a correlation, so it tells you the direction, not that the Overview alone did it. Seer's earlier 2025 study, also over a year old and possibly dated, found Overviews tend to land on queries that already had low click-through.

And on October 29, 2026, on Reddit's earnings call, CEO Steve Huffman said this about Google's AI Mode, which is the separate chat-style search, not the Overview box, per a third-party transcript of the call (The Verge reported it in near-identical words): "The reality is the AI mode does not drive traffic, not for us and not to anybody as far as I know. And so it's just a fundamental shift in how Google works. It used to be a traffic driver. Now it's not." Reddit's own search referrals are his data. The "to anybody" part is his opinion, and none of the studies above measured AI Mode directly.

Now the how. Google's list for AI features is the plain SEO list, and I'd walk it in this order because the first items are the ones that quietly knock you out. Check that robots.txt allows crawling, and also that your CDN or host isn't blocking Googlebot, because a firewall setting can do that without anyone on your team touching robots.txt. Then verify the site in Search Console and confirm the page is actually indexed and shows a snippet, since an unindexed page cannot be a supporting link no matter how good it is (we wrote up the usual reasons for why Google can't see a website). Then make sure the answer itself is in text on the page, not trapped in an image or a PDF, and that any structured data you have matches what a visitor can see. Link to the page from your other pages so it can be found. If you're a local business, keep your Business Profile current, and if you sell products, keep Merchant Center current, because Google names both.

Google also says Overviews may use "query fan-out," running several related searches across subtopics to build one answer. My read of that, and it is a read, not a promise from Google, is that a page answering the follow-up questions plainly has more ways to be picked up than a page that only targets the head phrase.

There are tradeoffs. If you don't want Google showing your text in these features, the controls are nosnippet, data-nosnippet, max-snippet, or noindex, and each of them also limits how you show up in regular Search, so blocking the Overview means giving up snippet visibility too. Changes also take time to register, since Google says recrawling can take from several days to several months.

The failure modes are mostly people buying fixes Google says it doesn't need. Somebody sells you an "AI file" or special schema for Overviews, and Google's page says you don't need new machine-readable files or special schema.org markup to appear. Or you do everything right and still don't get cited, because Google says meeting every requirement doesn't guarantee crawling, indexing, or serving, and that Overviews "often don't trigger" at all.

What to verify on your own site is simple and boring. Overview and AI Mode clicks are counted inside the normal Performance report in Search Console under the "Web" search type, so you are watching total clicks and impressions for the queries you care about and checking the pages by hand. Google says clicks from pages with Overviews tend to be higher quality, meaning people stay longer, and that is a claim worth checking in your own analytics rather than taking on trust.

The same question comes up for ChatGPT, Claude, and Perplexity, and the honest answer there is thinner, because none of the vendor pages we read (OpenAI's search help and crawler docs, Anthropic's crawler and web-search pages, Perplexity's crawler docs) explains how sources are chosen; OpenAI's one line is that ChatGPT "ranks search results using multiple factors intended to help users find relevant, reliable information." The same page says "placement is not guaranteed," and that to be eligible a site should allow OAI-SearchBot to crawl it and make sure the host or CDN lets in traffic from OpenAI's published searchbot IP addresses. OpenAI's crawler docs add that a site blocking OAI-SearchBot won't be shown in ChatGPT search answers (it can still show up as a navigational link), that blocking GPTBot is a separate setting about model training, and that robots.txt changes take about 24 hours to register for search. A separate agent, ChatGPT-User, visits pages when someone asks ChatGPT a question, and OpenAI says robots.txt rules "may not apply" to it and that it isn't used to decide what appears in search.

Anthropic's help center lists three crawlers for Claude, and the one that matters here is Claude-SearchBot, which it says indexes content to improve search results, so blocking it "may reduce your site's visibility" in Claude's search answers. Blocking Claude-User, which fetches pages when someone asks Claude a question, can cut visibility too, and blocking ClaudeBot is the training opt-out. Anthropic also says blocking its IP addresses may not reliably opt you out, so robots.txt is the switch to use. Claude's web search page says every search response includes citations with source links, but it doesn't say how Claude chooses among sources.

Perplexity's docs say PerplexityBot exists to "surface and link websites in search results," recommend allowing it in robots.txt and letting its published IP ranges through your firewall, and warn that Cloudflare's Bot Fight Mode can still challenge it even with an allow rule. Changes can take up to 24 hours. Perplexity's crawler docs don't describe how it ranks sources.

So for all three, the documented part is access, and it's the same shape as Google's first check: let the search crawler in through robots.txt and through your CDN or firewall, which is a separate switch from the training crawler, and you can block training while staying visible in search. Anyone selling you a ranking formula for these tools is selling something none of these vendors' help or crawler pages describes.

Here is what automation won't fix. A tool can generate pages fast, but a page that says what every other page says gives Google nothing to cite, and Google's own guidance keeps pointing back to helpful, reliable, people-first content. Someone who actually knows the answer still has to write it and check it, which is the same argument we made about whether a human should review AI-written content before it goes live.

A hypothetical to make it concrete: a bakery that writes "our gluten-free loaves are baked Tuesday and Friday, order by noon the day before" in plain text on an indexed page has something an Overview can quote. A bakery with the same fact inside a photo of a chalkboard does not.

Can I pay or add markup to get cited in an AI Overview? No. Google says there is no special schema, file, or optimization required, and the page only needs to be indexed and eligible for a snippet.

How much traffic does a citation bring? In Seer's 2026 data, cited brands saw 2.07% vs 0.94% on informational Overview queries in 2025, so roughly double (closer to 1.3x without one outlier account), on a smaller base than a normal results page.

Where do I see AI Overview traffic? In Search Console's Performance report under the "Web" search type, mixed in with the rest of your search traffic.

How do I get cited by ChatGPT, Claude, or Perplexity? Allow their search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot) in robots.txt and in your CDN or firewall. Beyond that, their help and crawler pages don't explain how they choose sources.

Does this apply to AI Mode too? Google gives the same eligibility rules for AI Overviews and AI Mode, but the two may use different models and show different links, and the click studies here measured Overviews, not AI Mode.

So picture two pages answering the same question. One is the chalkboard photo, nice to look at and invisible to the box at the top of the results. The other is two plain sentences of text on an indexed page, the only one of the two that box can quote.