Best Lists Rule AI Citations — But Self-Promotion Backfires Unless the Fit Is Right

Key takeaway: The content format AI answers cite most is the "best [category]" listicle. When Ahrefs analyzed 26,283 sources cited by ChatGPT, this format made up 43.8% of them (Allsopp study). But dropping your own brand into a listicle does not automatically work. In a follow-up experiment from the same company, 43% of the answers that cited a self-promotional page never mentioned the brand (Makosiewicz experiment). Self-promotion works only for narrow, well-fitting queries, and being cited does not mean the brand is seen (as of 2026-07-06).

---

Three-line summary

• The most-cited format is the "best" listicle. 43.8% of ChatGPT's cited sources were this format, and 79.1% of the cited listicles had been updated within 2025 — freshness is close to a precondition for citation.
• Self-promotion works only conditionally. It appeared in 66.4% of responses to a narrow, well-fitting query but dropped to 15.8% for a broad one, and 43% of answers that cited the promotional page never named the brand.
• Citation is a gateway, not the goal. Being cited does not mean users see your brand name, so even with a listicle tactic you have to measure "mentioned," not just "cited," to know the real effect.

"Best" listicles dominate AI citations — the 43.8% number

The content AI pulls most often when composing an answer is the "best [category]" listicle. When Ahrefs' Glen Allsopp classified 26,283 source URLs cited by ChatGPT across 750 top-of-funnel prompts (software, product, and agency recommendations), 43.8% of all cited sources were "best of" listicles — list-style posts like "best project management tools" or "top 10 CRMs" took the single largest share of any format.

Two supporting numbers sharpen the picture. 79.1% of the cited listicles had been updated within 2025, and 35% came from low-authority domains. The first says freshness is effectively a condition for being cited; the second says domain authority is not an absolute one. An old post loses its slot even if the domain is strong, while a lesser-known domain can still be cited if it holds a current comparison list. For why citations fade over time, see AI citation decay and freshness.

That said, this figure is drawn primarily from ChatGPT. The study only notes that listicles were slightly more prominent in Google's AI Overviews and gives no separate figures for Gemini, Claude, or Perplexity, so generalizing to "44% on every engine" overreaches. Engines differ in how they cite, as documented in the ghost citation analysis.

The self-promotion paradox — cited, but the name is missing

At this point the conclusion looks simple: "So we'll make our own 'best' listicle and put our brand in it." But a follow-up experiment from the same company shows that instinct is only half right.

Ahrefs' Mateusz Makosiewicz placed 34 pages promoting the company's own brands (the new "Ahrefs Evolve" conference and the "Brand Radar" tool) across 5 domains and tracked 9,886 answers from ChatGPT, Gemini, Perplexity, and Copilot over four months. The key finding was that a self-promotional page being cited and the answer actually naming the brand are two separate events. Among answers that cited a page promoting the conference, 43% never uttered the name "Ahrefs Evolve" at all. The link was attached, but the brand was absent from the sentences a user actually reads.

The phenomenon itself — that citation and mention are different events — was already laid out in the ghost citation analysis. What this experiment adds is how that gap opens up for the specific tactic of self-promotion. The experiment sums up the condition this way:

"Self-promotional content seems to work best when the brand is both a natural fit for the query and plausible enough to belong in the existing consensus." — Mateusz Makosiewicz, Ahrefs

In other words, self-promotional content works best when the brand naturally fits the question and is plausible enough to belong inside the consensus that has already formed. Conversely, a brand forced into the list gets filtered out in a specific way: the AI references the page but does not speak the name. At that point self-promotion is not "zero exposure" but "a gift to someone else" — your page is cited while the answer recommends a different brand.

It only works with a narrow fit — 66.4% vs 15.8%

The "natural fit" condition showed up most sharply in query breadth. For the very same conference, a narrow, well-fitting query — "best SEO conferences 2026" — placed it in 66.4% of responses, while a broad query — "best marketing conferences 2026" — reached only 15.8%. More than a fourfold difference.

The read is this. An SEO conference run by an SEO tool company fits naturally inside the consensus for the query "SEO conferences." But in the broad arena of "marketing conferences," established events with scale and recognition already own the consensus, leaving little room for a new one to squeeze in. The broader the category, the more the existing consensus resists; the narrower and more specific the query, the more room there is for a newcomer.

The gap also varied by content type. Conference (event) pages went unmentioned in 43% of the answers that cited them, but for the company's own product pages the unmentioned rate was just 11%. A product already fits its category query strongly, so citation converted to mention far more often. In the end, what decided mention was less "what you promote" and more "how naturally that thing belongs to the query's consensus."

New brands and established brands need different playbooks

Even the same self-promotion worked through opposite paths depending on brand maturity. The brand-new conference, which nobody knew yet, entered through empty slots in answers — and here its own promotional pages were cited and produced mentions. With no third-party content for the AI to lean on, the company's own pages became the only available basis.

The already-known in-house tool was the reverse. 94% of its new mentions came from third-party content, not the company's own pages. Because reviews, comparisons, and lists covering the brand already existed in the market, the AI named it based on what others wrote rather than the vendor's promotional pages. In practice:

• New / unknown brands — with the consensus still empty, using your own content to claim the empty slots of narrow, well-fitting queries is viable. But aim at a broad category and you risk the 15.8% case: cited yet unmentioned.
• Established brands — the marginal value of self-promotional pages is low. Since most mentions come from third parties, managing how others treat you (whether you appear in comparison lists and reviews) is more efficient than producing more of your own landing pages.

So what should you do — the listicle tactic and measurement

To sum up: the "best" listicle is the most-cited format, but putting your own name in one does not guarantee a mention. Distilling the practical principles where the two experiments overlap:

1. Target narrow, well-fitting queries first. Aim at "best [specific segment]," not "best [broad category]." The 66.4% vs 15.8% gap came from query-brand fit, not content quality.
2. Keep it fresh. 79.1% of cited listicles were recently updated posts. List-style content that is made once and left alone falls out of citations, so refresh elements like pricing, versions, and rankings on a schedule.
3. Measure "mentioned," not just "cited." Resting on the fact that a self-promotional page was cited misses the 43%-unmentioned case. Check, for the same query, whether the brand is named in the answer body and how often it appears versus competitors.
4. For established brands, manage third-party content. If 94% of mentions come from others' posts, your position in comparison and review lists matters more than your own landing pages. This connects to the perspective in measuring AI shelf share.

On measurement, RanketAI's brand visibility analysis repeatedly measures, per engine, whether a brand is cited as a source and whether it is mentioned in the answer body across ChatGPT, Perplexity, and Gemini, and competitor comparison shows the surfacing pattern against rival brands for the same query. Tools such as Semrush AI Visibility and Ahrefs Brand Radar also track citations and mentions — the key is watching both axes separately with a tool that fits your market and language.

Frequently asked questions

Can we just make a "best" listicle and put our product at number one?

A listicle on your own domain that ranks your own product first risks being cited without a mention unless the query is a narrow fit — the drop to 15.8% for the broad query is exactly that case. Keep your own lists to narrow, specific segments, and pair them with earning a natural spot in third-party listicles for broader categories.

If we just get cited, doesn't the traffic follow?

A citation link can be a path to click traffic, but brand awareness is separate. If the name is not spoken in the answer body, users do not know which brand it is until they click. When sales or awareness is the goal, citation rate and mention rate have to be viewed separately — a split covered in detail in the ghost citation analysis.

We are already a known brand. Do we still need more of our own content?

In the experiment, 94% of an established brand's new mentions came from third-party content. Managing how the market's comparison and review lists treat your brand has higher marginal value than adding more of your own landing pages. Keep your own content narrowed to defending narrow, well-fitting queries.

Can we trust these numbers for a non-English market?

Both datasets are English-language experiments from a single vendor (Ahrefs), and the self-promotion experiment in particular is a self-study on the vendor's own brands, so the sample is skewed. Treat the structure (listicles dominate, citation ≠ mention, narrow fit wins) as reference, but verify whether the absolute values and direction hold for your queries by measuring against your own brand.

Related reading

• What is a ghost citation? 62% of AI citations end without a brand mention — why citation and mention are different metrics. The conceptual foundation for understanding the 43%-unmentioned rate of self-promotion.
• Why AI citations decay over time — citation decay and freshness — the freshness lens on why 79.1% of cited listicles were recently updated posts.
• How to measure AI shelf share — the perspective for established brands managing their share of appearances in third-party content.