Key Takeaways
- Backlinks and total site traffic show no positive relationship with whether AI engines cite a page, based on published correlation research
- Content depth signals — word count, sentence count, readability — correlate more strongly than any classic authority metric
- Only about 10% of pages cited by one AI engine get cited by another, so a win in one assistant doesn’t carry over (full engine-by-engine breakdown in this comparison of the best AI visibility tools)
- Product and service pages earn more AI citations than blog content in the data we’ve analysed
- Your business can rank first on Google and still be absent from the AI answer sitting above it
For twenty years the deal was straightforward. Rank on page one, get clicks, get customers. Businesses invested in links, keywords and site authority because the relationship between those inputs and that outcome was reasonably well understood.
That relationship has partly broken. AI-generated answers now sit above the blue links on Google, and separate assistants — ChatGPT, Perplexity, Gemini — bypass the results page entirely. The uncomfortable part for anyone who has spent money on SEO: the things that got you ranked are not the things that get you cited.
The correlation data is blunt about this
Search Engine Journal published a correlation analysis of which content factors line up with brand citations across ChatGPT, Google AI Overviews and Perplexity. The results run against most of what the SEO industry sells.
Total backlinks correlated at -0.025 on average. Total traffic at -0.030. Total keywords at -0.030. All three are slightly negative, which in practice means: no relationship.
What did correlate, weakly but positively, were depth signals. Word count came in at 0.130 on average, sentence count at 0.102, and domain rating at 0.090 — the last of which was driven almost entirely by ChatGPT, where it hit 0.161, while barely registering on AI Overviews at 0.034.
None of these are strong correlations. That’s the honest reading, and anyone telling you they’ve found the one factor that guarantees AI citation is overselling. But the direction matters: the metrics your SEO agency reports on monthly are not the metrics predicting whether an AI assistant recommends you.
Why a #1 ranking doesn’t transfer
Two mechanical reasons.
First, the engines don’t share sources. We analysed 596,723 prompts where at least two of the five major AI engines answered the same question. Only 10.2% of cited URLs were cited by more than one engine. At the domain level it was 17.9%. ChatGPT was the extreme case — 97.15% of the URLs it cites appear on no other engine in our data. The only pairing with meaningful agreement was Google AI Mode and Google AI Overviews at 11.55%, and those two share infrastructure.
Second, ranking and citation are different decisions made at different stages. Dejan AI’s grounding research ran one identical query through Google, OpenAI and Anthropic on the same day. Gemini cited all 7 pages it retrieved. GPT-5.5 retrieved 39 pages and cited 2. Claude cited 9 of 14. Being retrieved is not being cited, and the gap between the two varies enormously by system.
What actually seems to earn citations
From our own analysis of 383,421 citations across 391 buying-intent prompts between January and June 2026, three patterns held up.
1. Product and service pages outperform blog content. By format, product and feature descriptions took 28% of citations and reviews took 22%. Comparison articles took 14.4% and how-to guides 12.6%. For a local or mid-sized business, that means your services page is doing more work than your blog — and it’s usually the page nobody has updated in two years.
2. Third-party discussion carries real weight. Reddit was the single most-cited domain in the set at 4.85%, ahead of every company’s own website, rising to 7.33% on Google AI Mode. AI assistants treat practitioner conversation as evidence in a way they don’t treat marketing copy.
3. Citations concentrate. The top 10 domains took 25.3% of all citations and the top 100 took 67.7%, with roughly 6,250 other domains splitting the rest. This is not a long tail where consistent effort gets everyone a share.
What this changes for a small or mid-sized business
The practical shift is from optimising position to optimising for being quotable.
Write service pages that state specifics — what you do, who it’s for, what it costs, where you operate, and honestly, what you don’t do. Vague brochure copy gives an AI assistant nothing to lift. Pages that answer a question directly and early tend to get used; pages that build to a sales pitch tend not to.
Keep the factual detail current. AI answers shift as new information appears, and outdated pricing, hours or service areas are a reason to be passed over rather than a minor housekeeping issue.
Pay attention to what’s said about you elsewhere. Reviews and forum threads about your category are being read as evidence. You can participate in those honestly — under real names, with your business disclosed — but manufacturing them is both a platform-ban risk and, for endorsements, a legal one.
And check more than one assistant. Because engine overlap is so low, confirming you appear in ChatGPT tells you close to nothing about Google AI Overviews.
A twenty-minute check you can run this week
Write down five to ten questions a customer would genuinely ask before choosing a business like yours — the problem, not your brand name. Run each through ChatGPT, Google AI Overviews, Perplexity and Gemini. Record two things: whether you’re named, and which websites the answer cited as sources.
That second column is the valuable one. Those cited sources are what the assistant trusted enough to build its answer from. If they’re competitor service pages, a review site and a forum thread, and none of them reference you, you now know exactly where the gap sits — and you learned it for free.
This content is brought to you by the FingerLakes1.com Team. Support our mission by visiting www.patreon.com/fl1 or learn how you send us your local content here.