AI visibility: being cited, not just indexed

# AI visibility: being cited, not just indexed

Being indexed means existing somewhere. Being cited means existing in the answer. It is no longer the same battle.

The number that unsettles most SEO teams

Ahrefs analyzed 15,000 long-tail queries in August 2025, comparing URLs cited by ChatGPT, Perplexity and Copilot against Google and Bing rankings for the same queries. The result: only 12% of URLs cited by these AI assistants appear in Google's top 10. The number that actually stings is this one: 80% of citations don't rank anywhere, not even in the top 100.

In other words, a company can be perfectly indexed, perfectly ranked on its keywords, and still be structurally absent from the conversation generative AIs are having with the user. Indexing guarantees a spot on the shelf. It no longer guarantees a spot in the answer.

What the studies show, disagreements included

This calls for rigor, because the numbers diverge by methodology, and a serious company does not build strategy on a single figure chosen because it sounds dramatic.

Averi analyzed 680 million AI citations in March 2026: only 11% of domains cited by ChatGPT are also cited by Perplexity. An independent Passionfruit study, around the same period, confirms a similar figure, close to 12% overlap across three engines. A UK firm, Qwairy, analyzed 118,000 AI-generated answers across ChatGPT, Perplexity, Google AI Mode and Claude, reaching the same conclusion: 11% of cited domains appear across multiple platforms at once.

But not every study agrees on the scale of the phenomenon. A Semrush analysis cites a 90% overlap between ChatGPT citations and Google's top 10, while a PragoMedia study puts it at 14%. The difference almost always comes down to methodology: query type tested, analysis window, definition of what counts as citation.

What stays constant across every serious study, though, is the direction: the correlation between Google ranking and AI citation is weakening, not strengthening. Ahrefs documented this decline specifically for AI Overviews: 76% of citations came from the organic top 10 in mid-2025, versus 38% by early 2026 according to Ahrefs, and 17% according to a separate BrightEdge analysis.

Ranking opens one door. Citation opens another, and it is not always the same door.

Why indexing and citation follow two different logics

This is not a mystery, it is architecture. Each AI platform builds its answer from a different source pool, with a different retrieval mechanism. ChatGPT combines its training knowledge with occasional web search via Bing. Perplexity crawls the web continuously with its own bot, independent of Google. Claude relies mainly on its training data, with more cautious web retrieval. Google's AI Overviews still build on the existing Google index and Knowledge Graph, which explains why they keep the strongest correlation with classic organic ranking among all platforms studied.

The concrete result: a page perfectly optimized for Google can be invisible on Perplexity, and vice versa. 28.3% of ChatGPT's most-cited pages have zero organic visibility at all, according to Ahrefs. These are not poorly optimized pages. These are pages answering a selection logic that classic SEO does not measure.

Who wins instead, and why

A handful of domains capture a disproportionate share of citations. According to an analysis by The Digital Bloom, just five domains, Wikipedia, YouTube, Reddit, Google properties and Amazon, account for 38% of all AI citations across engines. Wikipedia is ChatGPT's most cited source. Reddit dominates on Perplexity, making up 46.7% of its top ten most-cited sources.

What these numbers describe is not an arbitrary preference for certain sites. It is a preference for independent validation. An AI choosing a reliable source systematically favors a third-party site that corroborates a claim over the brand's own site making that same claim. This is exactly the Validation layer of the Entity SEO doctrine: external signals confirming an entity genuinely exists carry more weight than what it says about itself.

The marketing edge: stop managing a channel that no longer exists

Here is the most common budget mistake right now: treating AI as a single channel with a single checklist. The data contradicts that directly. Strong presence on ChatGPT predicts almost nothing about visibility on Perplexity. Optimizing for one without measuring the other is like buying ads on a single platform and assuming the audience follows everywhere else automatically.

The right question for a marketing committee is no longer: are we visible in AI? It is: on which platform, for which query, and through which third-party source are we cited? Three different questions, three potentially different answers, one budget to allocate intelligently across them rather than betting everything on classic SEO and hoping it covers all of it.

How this gets measured

This is exactly what Astronaut's AI Visibility score is built to capture: real presence, measured separately per platform, inside generated answers, distinct from the classic organic ranking that remains the focus of the Entity Confidence score's Structure pillar. A high AI Visibility score without a high Entity Confidence score signals an entity that is well cited but poorly structured. The reverse signals an entity solid on paper but absent from the conversation. The two read together, never in isolation.

What this means this week

Three concrete checks. Ask ChatGPT, Perplexity and Claude directly a question a typical client would ask about the category, and note whether the brand shows up, on which platform, and through which source it gets mentioned. Identify the three third-party sources, a sector directory, a review platform, Wikipedia if relevant, that could legitimately speak about the entity, and prioritize updating those over producing more content on the owned site. Stop tracking one averaged AI ranking and start tracking a per-platform table, even a simple one, because an average hides exactly the variance these studies just demonstrated.

What still holds true

Google ranking has not become useless. It remains the widest entry door, and on certain platforms, AI Overviews especially, the correlation stays real. But ranking alone no longer tells the full visibility story. A company measuring only its ranking is measuring an increasingly small fraction of where its customers actually go looking for an answer.

See how the AI Visibility score gets built inside Astronaut: Astronaut

Sources cited

• Ahrefs, analysis of 15,000 queries, August 2025, plus supplementary AI Overviews research, July 2025 and 2026 data.
• Averi, analysis of 680 million AI citations, March 2026.
• Passionfruit, cross-source overlap study across three AI engines, March 2026.
• Qwairy, analysis of 118,000 AI-generated answers, 2026.
• Semrush AI & SEO Report, 2026.
• PragoMedia, synthesis of Google/AI overlap data, March 2026.
• BrightEdge, separate AI Overviews analysis, cited early 2026.
• The Digital Bloom, domain concentration analysis, 2025.
• Profound, analysis of ChatGPT's most-cited sources, June 2025.