Every agency now sells AI search optimization, and most of the pitches share a flaw. They present the arrival of AI Overviews, AI Mode, ChatGPT and Perplexity as a clean break from search as it was, requiring a new discipline with a new acronym and a new retainer. The data does not support that. Some things have changed a great deal, and a marketing director who ignores them will watch organic traffic fall while reporting that rankings are fine. But the foundations that earn a citation from an AI answer are, to an uncomfortable degree, the same foundations that earned a first-page ranking in 2019.
This guide separates the two. It uses the largest studies published in the last eighteen months, from Google’s own documentation to clickstream panels covering tens of millions of searches, and it names the sources so you can check them. Where the studies disagree, the disagreement is stated rather than smoothed over.
The Numbers, Stated Honestly
Start with how often an AI answer appears at all, because the figure depends entirely on who is counting. Semrush, tracking more than ten million keywords, found AI Overviews on 15.7 percent of them in November 2025, down from a peak near 25 percent that summer. BrightEdge, tracking a curated set of commercial queries, puts presence at around 48 percent in February 2026. Similarweb’s clickstream data, which measures what people actually searched rather than a keyword list, says 43 percent of searches now show one, up from 15 percent a year earlier. The honest range is somewhere between one in six keywords and one in two real searches, and it is highest in the categories that matter to B2B marketers: BrightEdge measured 88 percent in healthcare and 82 percent in B2B technology by the end of 2025, and intent has shifted, with commercial and transactional queries making up roughly a third of AI Overview triggers by late 2025 compared with a tenth at the start of the year.
Then the click. Pew Research watched 900 US adults make nearly 69,000 searches and found that when an AI summary appeared, users clicked a traditional result on 8 percent of visits, against 15 percent without one, and clicked a link inside the summary on just 1 percent. Seer Interactive’s 2026 update across 53 brands and 5.5 million queries measured organic click-through at 2.36 percent with an AI Overview present and 3.82 percent without. Across the whole of US Google, SparkToro and Similarweb found that 68 percent of searches ended without a click in the first four months of 2026, up from 60 percent in 2024, and that clicks to the open web fell by almost 23 percent.
Ahrefs adds an important correction. Position-one click-through on keywords with an AI Overview fell 58 percent between December 2023 and December 2025, but click-through on informational keywords without one also fell, from 7.6 to 3.9 percent. Some of what gets blamed on AI is a broader change in how the results page is built. The direction is not in doubt. The attribution is messier than the headlines.
Finally, the assistants themselves. ChatGPT reported 900 million weekly users in February 2026. Similarweb counts generative AI platforms at 9.5 billion visits a month, up 70 percent in a year, but search engines still reach five times as many unique people, 95 percent of ChatGPT users also use Google, and AI referrals to external websites were flat between January 2025 and January 2026 even as AI usage grew by nearly a third. Only 6.8 percent of ChatGPT answers in May 2026 included any external link. As Similarweb put it, 93 out of every 100 ChatGPT answers send nobody anywhere. AI Mode passed a billion monthly users by Google’s count, and Google has said twice, most recently in June 2026, that it is not making AI Mode the default search experience.
What Changes
The Click Stops Being The Unit Of Success
For twenty years the purpose of ranking was the click, and everything downstream, from traffic reports to agency contracts, was built on it. That link is now broken for a large share of queries. Seer’s data shows that being cited inside an AI Overview earns roughly 120 percent more clicks per impression than not being cited, but still 38 percent fewer than the same query produced before the Overview existed. You can win the citation and still lose the traffic.
The practical consequence is that visibility and traffic must be measured separately. A page can be doing its job, answering the question that a buyer asks an AI assistant, shaping which brands appear in the answer, without a visit ever being recorded. Google acknowledged as much in August 2026 when it rolled out a Generative AI performance report in Search Console to every site: it shows impressions in AI Overviews and AI Mode, and deliberately does not show clicks, because for many of those impressions there is no click to show.
Citations Come From Well Beyond Page One
The old model assumed that if you ranked in the top ten you were in the conversation and if you did not you were not. AI answers do not work that way. Ahrefs’ March 2026 study of 863,000 result pages found that only 38 percent of AI Overview citations came from pages ranking in the top ten, with 31 percent ranking between eleven and one hundred and another 31 percent ranking outside the top hundred or not at all. BrightEdge, using a different query set, puts the top-ten overlap even lower, at about 17 percent. Peec AI’s analysis of 30 million sources found that 82 percent of AI Overview citations point to deep pages rather than homepages.
This cuts both ways. A page that has never ranked can be cited if it answers a specific sub-question better than the pages that do rank. And a page that ranks first can be passed over if the AI system finds a more direct answer elsewhere. Rankings remain a useful proxy, but they are no longer the gate.
Mentions Start To Matter As Much As Links
The strongest correlation study on AI visibility is Ahrefs’ analysis of 75,000 brands from May 2025. The factor most closely associated with appearing in AI Overviews was not backlinks, at a correlation of 0.22, or referring domains, at 0.30, or even Domain Rating, at 0.33. It was the number of times the brand was mentioned across the web, at 0.66, twice the strength of any link metric. Brands in the top quartile for web mentions averaged 169 AI Overview appearances against 14 for the next tier down, and a quarter of brands had none at all.
That is a shift in what off-site work is for. A mention in a trade publication, a listicle, an industry directory or a Reddit thread may carry no link and no measurable authority in a conventional SEO tool, and still be one of the signals that puts a brand into an AI answer. Semrush’s Ghost Citations study, published in June 2026, adds a twist: 62 percent of AI citations do not name the brand being cited, and the platforms differ wildly. ChatGPT links to sources 87 percent of the time but names the brand in only 21 percent of answers; Gemini names brands 84 percent of the time but links in only 21 percent. Comparative queries, the “best X for Y” kind, produced 2.4 times more brand mentions than informational ones.
Freshness And Specificity Are Rewarded
Ahrefs measured the age of nearly 17 million cited URLs and found that AI assistants cite content around 26 percent fresher than Google’s organic results. ChatGPT’s average cited page was 958 days old against 1,416 for a Google organic result, a gap of more than a year. Semrush found ChatGPT draws on roughly 15 sources per answer where Gemini uses about 3, and that only 36 brands managed to stay in the top 100 across ChatGPT, Gemini, AI Mode and AI Overviews every month for four months. Presence in AI answers is more volatile than a ranking, more sensitive to recency, and more rewarding of pages that answer one specific question well than of pages that cover a topic broadly.
Measurement Has To Be Rebuilt
None of the standard reporting was designed for this. In Search Console, clicks and impressions from AI Mode have been folded into the ordinary Web search type since June 2025 and cannot be separated; the new Generative AI report shows impressions only, combining Overviews and AI Mode. In GA4, a click from an AI Overview arrives as ordinary Google organic traffic and is invisible as AI traffic. Visits from ChatGPT, Perplexity, Gemini, Claude and Copilot arrive as referrals and need a custom channel group built on a source regex to be seen together, and clicks from the mobile apps often land as direct traffic. Third-party tools such as Ahrefs Brand Radar and Semrush’s AI toolkit exist to track citations and mentions across platforms, and they are the only way to see what is happening on ChatGPT or Perplexity, since neither publishes a console. In Semrush’s 2026 survey, 45 percent of marketing leaders said they could not measure their visibility in AI answers at all.
What Does Not Change
You Still Need To Be Crawled And Indexed, By The Same Crawler
Google’s documentation on AI features, updated in December 2025, states the eligibility requirement in one sentence: a page must be indexed and eligible to appear as a snippet. AI Overviews and AI Mode are populated by Googlebot, the same crawler that has always fed search. There is no separate AI index to submit to. The controls that keep a page out of AI features are the ones that already existed, nosnippet, data-nosnippet, max-snippet and noindex, and each removes you from ordinary results too. The Google-Extended directive, which many sites added in 2023, governs only whether content trains future Gemini models and grounds the Gemini app; Google states plainly that it does not affect inclusion in Search or AI Overviews.
The other platforms have converged on the same three-crawler pattern. OpenAI uses GPTBot for training, OAI-SearchBot for the ChatGPT search index and ChatGPT-User for pages fetched on a user’s request. Anthropic uses ClaudeBot, Claude-SearchBot and Claude-User in the same roles. Perplexity uses PerplexityBot for its index and Perplexity-User for on-demand fetches, and notes that the latter generally ignores robots.txt because a person asked for the page. Blocking the training crawler, which many publishers did in 2023, does not remove a site from AI search answers. Blocking the search crawler does, and Anthropic’s documentation says so explicitly. A surprising number of sites are still carrying blanket blocks written before this distinction existed.
There Is No Special File, Markup Or Format
Google’s guide to optimising for generative AI search, updated in July 2026, could not be clearer. There are no additional requirements to appear in AI Overviews or AI Mode. You do not need to create new machine-readable files, AI text files, markup or Markdown, because Google Search does not use them. There is no special schema.org structured data to add. John Mueller said of the llms.txt file in June 2026 that it is purely speculative for now, that the file has existed for years and none of the AI systems use it, and that the most basic form of optimising for agents is simply not blocking them. Neither OpenAI nor Anthropic has said it reads llms.txt.
Structured data deserves a specific note because it is the most commonly sold AI optimization tactic. Ahrefs tested it directly, comparing 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 control pages. Adding schema produced no meaningful uplift in citations on any platform. On AI Overviews the cited group actually moved down by 4.6 percent, a statistically significant result. Schema still does what it always did, which is help Google understand entities and populate rich results. It does not buy AI citations.
Content Quality Is Still The Whole Game
Strip away the vocabulary and Google’s advice on AI search is the advice it has given for a decade: foundational SEO, unique content that is not a commodity, and accurate business information in Business Profile and Merchant Center. The citation studies say the same thing from the other direction. AI systems cite the page that answers the question most directly, most recently and with the most specific detail. That is the page a competent editor would have told you to write anyway. What has changed is the penalty for not doing it, because a mediocre page that used to earn a position-six click now earns nothing.
Authority Still Compounds, It Is Just Counted Differently
Links have not stopped mattering. In Ahrefs’ brand study, referring domains and Domain Rating still correlated positively with AI visibility. They have been joined by mentions, and the two reinforce each other: a link from a trade publication is also a mention, a listicle placement is also a citation source, and a brand that appears consistently across Reddit, YouTube, LinkedIn, Wikipedia and the trade press, which Peec found to be the most-cited platforms overall, is building the same reputation that used to be expressed only as link equity. The work of being talked about by credible third parties is older than Google. It has simply become easier to measure and harder to fake.
The Buyer Still Exists, And Converts Better When They Arrive
The last thing that has not changed is that a person with a problem still ends up on a website before they buy. The volume has shifted, but the value of each arrival has risen. Semrush measured AI search visitors as 4.4 times more valuable than organic visitors by conversion rate in 2025. Ahrefs found AI search brought 0.5 percent of its visitors and 12 percent of its signups. Similarweb’s January 2026 data shows ChatGPT referrals spending fifteen minutes on site across twelve pages with a 7 percent conversion rate, against eight minutes, nine pages and 5 percent from Google. Part of this is selection: the person who clicks out of an AI answer has already read the summary and decided they need more. But it means the page they land on carries more weight than it used to, and 59 percent of that AI referral traffic, per Similarweb, lands on the homepage.
What To Actually Do
The practical programme that falls out of this is shorter than most agencies would like. First, audit robots.txt and remove any blanket blocks on AI crawlers that were written before search and training crawlers were separated; keep the training blocks if you want them, but let OAI-SearchBot, Claude-SearchBot and PerplexityBot through, and never touch Googlebot. Second, rebuild reporting so that AI impressions in Search Console, AI referrals in GA4 and citations in a third-party tracker sit alongside rankings and traffic rather than being hidden inside them, and reset the conversation with leadership about what success looks like before the traffic line makes it for you. Third, look at the deep pages that answer specific questions in your category and make each one more direct, more current and more specific, because those are the pages that get cited from outside the top ten. Fourth, treat mentions as a channel: pursue placements in the trade publications, comparison lists and communities where your buyers and the AI systems both look, and check whether those placements name the brand and not just the URL. Fifth, make the homepage worth landing on, since that is where most AI referrals arrive. And stop paying for anything sold as AI-specific markup, files or formats, because Google has said in writing that it does not use them and Ahrefs has shown that they do not move citations.
HighFlyers’ SEO and link building service is built around that sequence, audit first, then on-page depth, then the outreach that earns mentions. For a closer look at the mention side, see the REVERB guide to what a brand mention is for SEO.
Conclusion
AI search optimization is real, and it is not new. What has changed is the economics of the click, the reach of citations beyond page one, the weight of unlinked mentions, the premium on fresh and specific pages, and the reporting needed to see any of it. What has not changed is the requirement to be crawled and indexed by the same crawlers as before, the absence of any special file or markup that buys inclusion, the primacy of content that answers the question better than anyone else, and the slow compounding of authority earned from credible third parties. A marketing director who holds those two lists side by side will make better decisions than one who buys the acronym.
Sources
Google Search Central, AI features and your website and Optimizing for generative AI search; Google, Generative AI performance report; Semrush AI Overviews study; BrightEdge, AI Overviews at the one-year mark; Pew Research on AI summaries and clicks; Seer Interactive CTR update 2026; SparkToro and Similarweb zero-click study 2026; Ahrefs on AI Overviews and clicks; Ahrefs on citations and the top ten; Ahrefs brand correlation study; Ahrefs on citation freshness; Ahrefs schema and citations test; Semrush Ghost Citations study; Semrush AI Visibility Index 2026; Similarweb AI search statistics 2026; Peec AI citation study; John Mueller on llms.txt; Google common crawlers; Perplexity crawlers; Anthropic crawler documentation via Search Engine Land; Orbit Media on tracking AI traffic in GA4.
