Your own "best X" list gets cited by AI. Does it get you recommended?

KEY TAKEAWAYS

1. Track three levels per prompt: retrieved, cited and recommended. Most dashboards stop at cited.
2. Self-ranked lists helped a new, niche-fit brand and did little for an established one.
3. Engines differ. Copilot moved fast in one test, ChatGPT did not move at all.
4. Third-party lists drove most new mentions in every study. Spend there first.

A "best X software" page that puts its own product at number one used to be a standard play for AI visibility. The logic was simple. AI engines love lists, so publish one and get cited.

The 2026 data says the citation part still works more often than not. The recommendation part often does not. Several studies now show the same pattern: the engine reads your list, cites it, and then names a competitor. This piece sets out what the evidence supports, where the studies disagree, and a way to decide whether your own list is helping.

Why is a citation not the same as a recommendation?

An AI answer to "what is the best project management tool" does three separate things with a page. It can retrieve the page while searching. It can cite the page as a source. It can recommend a brand in the answer text. These are different outcomes, and most visibility reports only track the second.

Ahrefs ran the cleanest test of the gap. Between 7 February and 31 May 2026 it tracked 34 self-promotional pages across five domains, promoting two of its own products: the Ahrefs Evolve conference and the Brand Radar tool. It analysed 9,886 answers from ChatGPT, Gemini, Perplexity and Copilot.

For the conference, 43% of answers that cited one of the pages still did not name Evolve. Among answers where a page was found but not cited, 74% skipped Evolve. For the tool, the same figures were 11% and 15%. The lesson is that being in the source list does not put you in the answer.

Lily Ray's study of Google AI Overviews makes the same point at larger scale. Using 100 B2B "best [category] software" queries, captured on 15 April, 15 May and 8 June 2026, she found that in 224 of 323 cited self-promotional listicles (69%), the publishing brand was left out of the recommendation. Eighty of the 100 queries showed an AI Overview.

Is it getting worse?

Ray reran the same 100-query set in September 2026. AI Overviews cited 38% fewer self-promotional listicles than in June. When they did cite one, the publisher was left out of the recommendation 83% of the time, up from 69%. In 70% of cases the list helped a competitor get recommended.

Two cautions apply. The September post documents its method less fully than the June study. And one 100-query B2B set is a narrow sample. Treat the direction as real and the exact size as uncertain.

The policy context points the same way. Google's spam policies, last updated 28 August 2026, now count as spam "attempting to manipulate generative AI responses in Google Search." Search Engine Roundtable traced that clause to mid-May 2026. Google's AI optimization guide adds that seeking inauthentic mentions "isn't as helpful as it might seem," and describes generic tip lists as commodity content.

Do self-ranked lists help anyone?

Yes, a narrow group. The Ahrefs experiment split cleanly by brand stage.

For Evolve, a new event with little existing coverage, the pages filled 72 answer slots that had previously named nothing, and 82% of those answers cited one of the pages. For Brand Radar, part of a well-known brand, only 6% of new mentions came from the self-promotional pages. The other 94% came from third-party content.

Query fit mattered too. Evolve appeared in 66.4% of answers to "best SEO conferences 2026" and only 15.8% for "best marketing conferences 2026". A self-ranked list works best where the brand is a natural answer to a narrow question and few other sources exist.

Scrunch's study, reported by PPC Land, reached a similar conclusion. It looked at about 10,000 AI-cited URLs across seven engines, with responses collected from 17 May to 30 June 2026, and compared self-ranking listicles with 1,033 neutral third-party comparisons. When a brand's own list was cited, its recommendation rate rose from about 4% to about 7%, and its mention rate from about 19% to about 39%. But 63.9% of answers citing a listicle recommended nobody, and a competitor was recommended instead of the author in 24.3%.

Scrunch also ran placebo checks. Only one scenario survived: a lower-visibility brand naming lower-visibility rivals, with a lift of 7 points against a placebo of minus 1.9. The causal estimates used a capped sample of about 6% of the eligible population, so the effect sizes are rough.

Does the engine change the answer?

A lot. In the Ahrefs test, Evolve's share of Copilot answers went from close to zero to 39% in March, 57% in April and 65% in May. Gemini picked it up less steadily. Perplexity barely moved. ChatGPT never named it for the tracked query.

Scrunch measured how often each engine recommended a brand when citing a neutral roundup. The spread is wide:

The same list can be a recommendation engine on Gemini and a reference footnote on Perplexity. Any decision about listicles has to be made engine by engine.

Did GPT-5.6 halve listicle citations in ChatGPT?

This is where the sources disagree most.

Peec AI compared a week before and a week after the GPT-5.6 rollout across about 1 million tracked prompts and 180 million sources. Listicles fell from 15.77% to 7.80% of ChatGPT citations. Comparison pages fell from 9.08% to 6.17%. Many posts reported this as listicles losing half their value.

Two other findings qualify it. First, cited sources per chat rose from 12.48 to 25.55 over the same period. SEO Handbook worked out that listicles per chat stayed about flat, at roughly 1.97 before and 1.99 after. Share halved because the pool doubled. Second, Nicolas Sitter tracked 616 hotel prompts weekly and found listicle share rose from 33.8% to 36.4% across the same switch, using his own title and URL rules.

The best reading of all three: GPT-5.6 diluted listicles rather than removed them, and the effect depends on the category. Software lists, where self-ranking is common, fell. Travel lists, where the format is natural, did not. Peec's own earlier study had self-promotional lists at about 3.6% of ChatGPT citations, against 10.3% for AI Mode and 10.4% for Perplexity, so ChatGPT was already a hard engine for this tactic.

What do the top guides on this miss?

Most pages ranking for this topic ask "do listicles work in AI search" and answer with a yes or no. Four things are usually missing.

1. Three levels, not one. They count citations. The studies above show recommendation is the outcome that matters, and it often moves the other way.
2. Brand stage. The 82% versus 6% split in the Ahrefs test is the single most useful number for deciding, and it rarely appears.
3. Engine differences. Copilot and ChatGPT reacted in opposite ways to the same pages.
4. Decay. In the Ahrefs test, pages were cited on about one in three eligible days, and the longest unbroken run for a page, engine and query was 52 days. A list is not a permanent position.

Should you keep, rewrite or retire your own list?

Use your brand stage and the query fit to decide. This table summarises what the evidence supports.

How do you measure whether your list is working?

A short method you can run monthly.

1. Pick the prompts. Choose 10 to 20 "best X" and "X alternatives" prompts your list targets. For sample sizes, see how to measure AI visibility when answers change.
2. Run them per engine. Record each engine separately. A blended number hides the Copilot and ChatGPT split.
3. Log three fields per answer. Was your page retrieved, was it cited, was your brand recommended in the answer text.
4. Add a counterfactual. Track the same prompts for a rival without a self-ranked list, or compare against the weeks before your page went live. Scrunch's placebo checks cut its apparent effects by a quarter to a third, so an uncontrolled before-and-after will overstate the lift.
5. Decide on recommendation rate. If your list is cited but your recommendation rate is flat or falling while rivals gain, the page is working for them. Rewrite or retire it.

Where ChatGPT now runs searches scoped to brand domains, a fair comparison page on your own site can still be read as a primary source. That is a different job from ranking yourself first; see the ChatGPT site: operator shift.

Where should the effort go instead?

Every study here agrees on one point. Third-party lists carry more weight than your own. Peec AI's study of third-party listicles, published in May 2026 and covering nearly 200,000 responses across eight engines, found that ranking first on a frequently cited list went with a visibility lift of 16.5 percentage points in B2B SaaS. That study was observational, so it shows association, not cause. Ahrefs' established product got 94% of new mentions from other people's pages.

In digital authority terms, a self-ranked list is a weak signal for both the press coverage pillar and the AI visibility pillar. A place on an independent comparison that engines already cite moves both. If you want to find which third-party lists engines cite in your category, the placement marketplace is built around that question.

Questions

Do AI engines still cite self-promotional listicles?

Yes, but less than they did. Peec AI found about 11% of software-review citations came from self-promotional lists between December 2025 and February 2026. Lily Ray's 100-query B2B set showed Google AI Overviews citing 38% fewer of them in September 2026 than in June.

Why would a cited listicle help a competitor?

The engine reads the page as a source of candidates, not as a verdict. In Lily Ray's September data the publishing brand was left out of the recommendation 83% of the time, and in 70% of cases the list helped a competitor get recommended instead.

Did GPT-5.6 stop ChatGPT citing listicles?

Not in absolute terms. Peec AI data showed listicles falling from 15.77% to 7.80% of ChatGPT citations, but sources per chat roughly doubled, so the count per chat stayed close to flat. A hotel-search study found listicle share rose slightly over the same switch.

Is ranking yourself first against Google's policies?

A neutral comparison is not spam. Google's spam policies, updated in 2026, now include attempting to manipulate generative AI responses in Google Search, and its AI optimization guide warns that inauthentic mentions are not as helpful as they seem. A self-ranked list built only to steer AI answers is the risk.

Sources

1. Ahrefs, Self-promotional content works until it backfires (AI SEO experiment) ahrefs.com
2. Lily Ray, Why calling yourself the best could be helping your competitors win in AI search lilyraynyc.substack.com
3. Lily Ray, Prediction. The next massive Google update is around the corner lilyraynyc.substack.com
4. PPC Land, Self-ranked brands gain AI recommendations from 4% to 7%, Scrunch finds ppc.land
5. Peec AI, Self-promotional listicles analysis. Data from 232,000 citations peec.ai
6. Peec AI, The listicle rank effect. What nearly 200,000 AI responses across 8 AI engines reveal peec.ai
7. SEO Handbook, ChatGPT now cites twice as many sources, and listicles lost half their share seohandbook.co.uk
8. Nicolas Sitter, Did GPT-5.6 really kill listicles? Not in hotel search nicolassitter.com
9. Google Search Central, Spam policies for Google web search developers.google.com
10. Google Search Central, AI optimization guide developers.google.com
11. Search Engine Roundtable, Google says spam policies apply to AI responses seroundtable.com

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →