Guide 06 · Evidence review
Do backlinks drive AI search visibility?
They correlate, weakly, and they are the weakest of the three signals usually measured together. Backlinks sit at 0.22. Unlinked brand mentions sit at 0.66. YouTube presence sits at 0.74. If you have one budget and you spend it all on links, you are buying the cheapest-correlating of the three.
Why would a link matter less here than in classic search?
Because the mechanism is different, and the difference is easy to state.
A classic search engine ranks documents, and a link is a vote that helps decide the order. An answer engine retrieves a handful of pages and writes prose from their text. What ends up in the answer is what was in the text it read. An unlinked mention of your plugin in a listicle puts your name in that text just as effectively as a linked one. The link is a bonus. The name is the payload.
That reframes the whole activity. You are not accumulating authority; you are trying to be present in the pages the engine happens to read. Those are different jobs and they reward different work.
What does that look like in a real answer?
Here is the concrete version, from ten WordPress plugin categories put to both Perplexity and Google AI Mode:
| Where answers come from | Perplexity | Google AI Mode |
|---|---|---|
| Share of sources owned by a product vendor | 41% | 47% |
| Categories citing youtube.com | 0 of 10 | 8 of 10 |
| Source overlap between the two engines | 15% (median) | |
Roughly half of every answer comes from pages owned by companies selling products in the space — writing, in most cases, about categories they do not sell in. Those are not high-authority independent publishers. They are vendors with a blog, which is a position you already occupy.
One publisher network in that data was cited on Perplexity in seven of ten categories and on Google in eight, and in six or seven of those it sells nothing at all. Whatever earns that is not a link profile. It is a large volume of long, structured, comparison-shaped articles about neighbouring categories.
0.74 is the strongest correlation on the list, and our own data shows why it is not evenly distributed: Google AI Mode cited youtube.com in eight of ten categories and Perplexity in none. If you are absent from Google’s answers specifically and you have no video, that is the most concrete gap in this dataset. No amount of writing substitutes for it there.
The 0.66 for mentions has started a market in bought and seeded mentions. Google’s AI optimization guide says seeking inauthentic mentions is not as helpful as it might seem, because its AI features depend on the same spam systems as the rest of Search. The evidence is on its own page: Do manufactured brand mentions work for AI search?
So should you stop building links?
No, and it would be a silly conclusion to draw from a 0.22. Links still do their classic-search job, they still bring referral traffic, and the pages that mention you often link to you anyway — so a mention strategy tends to produce links as a by-product while the reverse is not true.
The claim is narrower and it is about ordering. If you are choosing between a month of link outreach and a month of getting your product named in the roundups your buyers read, the evidence points at the second. And a paid link that carries no editorial mention is the worst version of this trade: you have bought the weakest signal and skipped the strongest.
| If you have one month | What it buys | Signal |
|---|---|---|
| Get named in category roundups | Your name in the text engines read | 0.66 |
| Publish a review or walkthrough video | Eligibility on Google’s answers specifically | 0.74 |
| Write about neighbouring categories | The pattern that travels across both engines | Observed |
| Buy links with no mention | A vote in a different election | 0.22 |
How much should you trust these numbers?
All three figures are correlations from observational studies of what happens to be cited. They cannot separate cause from a common cause, and the three run together — brands that get mentioned also tend to get linked and to have video. Nobody has isolated the mechanism.
A well-known brand generates mentions, links and videos all at once, and is also more likely to be cited for reasons that have nothing to do with any of them. Every one of these correlations is consistent with “being big causes all four”.
What survives that objection is the ordering, not the decimals. Under any reading, links are the weakest of the three, and the gap between 0.22 and 0.66 is too large to be an artefact of measurement. That is a weaker claim than a causal one, and it is the only one the evidence supports.
Published work in this area also leans heavily toward large, already-visible brands, which is not the situation a plugin founder is in. Treat the ordering as a guide to where the cheap wins are not, rather than as a formula.
How this was made: figures come from our own runs in the engines’ public interfaces or from the studies named above. Drafted with an AI assistant (Claude) and reviewed by Alauddin Chowdhury. Method and AI use.
Who is being read in your category?
Send me your wordpress.org slug. I will pull the full source list for your category from both engines and tell you which pages the answer is actually built from — including which ones name your competitors and not you. Free, no call, usually within a day.
Change log, 4 October 2026: backlink correlation corrected from 0.19 to 0.22, the figure Ahrefs published in its May 2025 study of 75,000 brands.