Ahrefs ran the experiment on itself, and it picked its two subjects carefully. Between February 7 and May 31, 2026 it tracked 34 of its own promotional pages across 9,886 answers from ChatGPT, Gemini, Perplexity and Copilot, testing a brand the engines had never mentioned (its new Evolve conference) against one they already cited routinely (its Brand Radar product). Both sides were measured the same way: when a brand newly appeared in an answer, what did the engine cite to get there? For the unknown brand, Ahrefs' own pages. For the established one, almost entirely other people's. The study published July 6, 2026, and that contrast is the finding: it puts a number on where owned-content spend buys AI visibility and where it buys nothing.
Credibility: MEDIUM. Transparent method and disclosed sample sizes, but a single vendor source with a clear self-interest: Ahrefs ran the test on its own two brands and publishes to advance its AI-visibility tooling. No independent corroboration; treat the direction as sound and the exact percentages as one lab's result.
- Unknown brand (the Evolve conference): of the answers where Evolve newly appeared, 82% cited one of Ahrefs' own promotional pages as the source. Evolve turned up in 66.4% of answers to "best SEO conferences 2026", and Copilot mentions climbed from near-zero to 65% by May 2026.
- Established brand (Brand Radar): of the answers where Brand Radar newly appeared, only 6% cited an Ahrefs promotional page. The other 94% cited someone else's content.
- The backfire: 43% of the answers that cited an Evolve promotional page never named Evolve anywhere in the answer. The engine took the page, kept the list, dropped the brand. For Brand Radar the same figure was 11%.
- Durability: cited pages showed up on only about one eligible day in three, and in Google's AI Overviews nearly half the cited sources were new each time.
Stay ahead of AI search
The changes that move your rankings and AI citations — and the exact move to make — before they cost you. Free, three times a week.
Start with the 43%. Nearly half the answers that cited the Evolve promotional pages went on to describe the category without ever naming Evolve: the engine took the page, kept the list, dropped the brand. A citation is not a mention, and self-promotion reliably buys only the first.
For a marketing lead deciding where next quarter's content money goes, one question settles it: does the AI answer already name us? If the engines have never mentioned the brand, self-promotional listicles and reviews are a legitimate way to seed the answer. 82% of the answers that newly named Evolve got there by citing an Ahrefs page. If the brand is already named, more owned promo pages are close to a rounding error: 94% of Brand Radar's new mentions came from someone else's content. Ahrefs also has an interest in you believing AI visibility can be engineered, so treat the exact percentages as one lab's result. The direction is harder to argue with.
- Pull your ten highest-intent queries, ask them in ChatGPT and Perplexity this week, and sort them into two piles (brand named, brand absent) before a dollar of content budget is allocated.
- Where the brand is absent, publish or refresh self-promotional listicles and reviews on high-authority domains. Then re-ask the query and confirm the answer names the brand, not just the page.
- Where the brand is already named, move budget from owned promo pages to third-party placements (earned reviews, analyst lists, comparison sites) and write the rationale down, because owned pages contributed only 6% of new mentions here. Keep the visibility tracker running while you do it: cited pages showed up on roughly one eligible day in three, so one week's spot-check tells you about that week and not much else, and the tracker is what shows whether the mention is actually holding.
Ahrefs Blog: "Self-Promotional Content Works—Until It Backfires (AI SEO Experiment)", Mateusz Makosiewicz (reviewed by Ryan Law), Ahrefs, July 6, 2026 — https://ahrefs.com/blog/self-promotional-content-ai-seo-experiment/