Rivals rank themselves first in 'best tools' lists. Does it get them recommended by ChatGPT and AI Overviews?

Do 'best tools' lists get you recommended by ChatGPT?

Sometimes, and far less often than the rivals publishing them expect. AI engines cite "best tools" lists heavily, but citing a page and recommending its author are different events. The gap between the two is where most self-ranked lists fail.

Three studies published between December 2025 and September 2026 looked at the question from different angles. Ahrefs counted which page types ChatGPT cites. Lily Ray, writing in Search Engine Land, examined how Google's AI Overviews treat self-serving listicles and which brands those answers go on to name.³ Scrunch measured what happens to an author's recommendation rate when its own list is cited.

This post reconciles the three studies. It turns them into a decision for a B2B SaaS head of marketing: whether to write the list, whether to rank yourself first, and how to measure what the page did. The short version is that the page shape works and the self-ranking mostly does not. The rivals doing it are reading the citation study and skipping the recommendation one.

Two definitions and a disclosure

A disclosure before the evidence. Sigzen AI sells AI visibility audits and programmes at published prices, so we have an interest in you measuring this.¹² The decision below works the same whichever tool you use, including a spreadsheet.

Listicle

A page built as a ranked or numbered list, such as "the best invoicing tools for mid-sized firms". A self-ranked listicle is one where the publisher includes its own product, usually at the top.

Recommendation rate

The share of AI answers to a set of prompts that advise the buyer to choose a given brand, as distinct from merely mentioning it or citing a page about it.

Ahrefs: 'best X' lists are the page type ChatGPT cites most

Ahrefs, which sells SEO and AI visibility tools, found "best X" lists were the most prominent page type among ChatGPT's sources, including lists whose authors ranked themselves first, with the data period not stated.¹ The study, by Glen Allsopp and published on 4 December 2025, covered 26,283 source URLs cited for 750 top-of-the-funnel prompts about software, products and agencies; geography is not stated.¹

Across all source links, "best X" blog lists made up 43.8% of page types (period not stated).¹ Of the lists Ahrefs inspected, 67.6% ranked the publishing company number one, and 79.1% had been updated during 2025, with the period of the sample again not stated.¹

Ahrefs also reported a correlation between a brand ranking highly in third-party lists and appearing in responses more often, with the period not stated.¹ The study counts citations, so it tells you which pages ChatGPT reads. It says little about whether the author's product reached the buyer's shortlist.

Our view: this is the strongest evidence that the page shape works, and the weakest evidence that self-ranking does.

Three studies in ten months

1. 4 Dec 2025

   Ahrefs: 26,283 ChatGPT source URLs for 750 prompts; "best X" lists the most cited page type, period not stated.¹

2. Jul 2026

   Search Engine Land, Lily Ray: how Google's AI Overviews treat self-serving listicles and the rivals they go on to name.³

3. 24 Sep 2026

   Scrunch: 818 self-ranked listicles against 1,033 neutral comparison pages, cited 17 May to 30 Jun 2026.²

Being cited is not being recommended

A citation puts your page in the engine's hands; a recommendation puts your product on the buyer's shortlist. Scrunch, which sells AI visibility software, measured the distance between the two in a study published on 24 September 2026.² It is a vendor study, and of the three it is the one that measured recommendations against a control group.

The sample was roughly 10,000 AI-cited URLs collected between 17 May and 30 June 2026, across seven AI assistants, with geography not stated.² Scrunch compared 818 self-ranking listicles against 1,033 neutral third-party comparison pages as a control, over 5,434 page and response pairs.² When a brand's self-promoting listicle was cited, its recommendation rate rose from about 4% to roughly 7%.² That is a real lift, and a small one.

What the engine does with the page

In 63.9% of responses the assistant treated the listicle as raw material for a balanced overview rather than adopting its number-one ranking.² Mentions moved more than recommendations: about 39% of answers citing the listicle named the author brand, against about 19% when it was not cited.² The engine lifts the name from the page and then decides for itself who wins.

Scrunch also ran a placebo test and found that about a quarter to a third of the lift may reflect trends already under way before the citation.² Read the move from about 4% to roughly 7% as an upper bound, with a smaller true effect behind it.²

What a cited self-ranked listicle earns its author

Citation and mention

• Page cited by the assistant
• Author named in about 39% of those answers, against about 19% when not cited
• Ranking rebalanced in 63.9% of answers

Recommendation

• Author recommended in roughly 7% of answers, up from about 4%
• A rival recommended instead 24.3% of the time
• A quarter to a third of the lift explained by prior trend

What happens to a self-ranked list in Google's AI Overviews?

Google's AI Overviews cite self-serving listicles and often name a rival in the same answer. Lily Ray, writing in Search Engine Land in July 2026, examined self-serving listicles cited by AI Overviews and reported that the recommendation frequently went to a competitor rather than to the author.³

Scrunch's per-platform figures point the same way. Among the five assistants our AI answer monitor also measures, the author's recommendation rate when its listicle was cited was highest on Gemini, at 14.8%, in the 17 May to 30 June 2026 data.² It was 8.9% on Google AI Overviews, 8.6% on ChatGPT, 7.7% on Google AI Mode and 4.8% on Microsoft Copilot.² Across all seven assistants, a rival was recommended instead of the author 24.3% of the time.²

The buyer is already sceptical

The buyer on the other side is already wary. In G2's survey of 1,076 B2B decision makers in North America, EMEA and APAC (March 2026), 64% said they meet inaccurate AI recommendations often or very often.⁴ Often meant a few times a month; very often meant weekly or more. Comparing vendor strengths and weaknesses was the most common reason they gave for using AI chatbots in software research.⁴

That scepticism is the environment any list is read in. The buyer compares what the answer says with what they already know of the category, and a ranking that contradicts it costs the author credibility rather than earning it.

Our view: a list that overclaims hands the engine a reason to balance it, and hands the buyer a reason to distrust it. Google's engines rebalance more readily than Gemini does, on these numbers, and Copilot barely moves.²

Author recommendation rate when its self-ranked listicle is cited, by assistant

Gemini14.8%

Google AI Overviews8.9%

ChatGPT8.6%

Google AI Mode7.7%

Microsoft Copilot4.8%

Why the engine lifts the page, not the brand

Engines retrieve several pages and compose one answer from them, so a single list is one voice among many. In Semrush's 2026 AI Visibility Index of 126 million US prompts (January to April 2026), ChatGPT cited about 15 sources per response and Gemini about 3.⁸ Your list is read alongside the others, and the ranking it proposes is checked against them.

Most of those citations go to company sites rather than to publishers. Profound, which sells AI visibility software, analysed 11.84 billion citations across eight engines between 16 April and 16 July 2026 and found about 57% went to company-operated sites.⁷ A category list on your own domain is therefore a plausible source. So is the one on your rival's domain, which is why both get cited and the engine arbitrates. Our guide to owned and third-party citations by engine sets a target for each.

Ranking is not the gate

A Google ranking is not the gate either. In an Ahrefs study of 863,000 search results and 4 million cited URLs (March 2026, period not stated), 38% of pages cited in AI Overviews ranked in Google's top 10 for that query.⁹ That is a statement about citations, not about rankings. A list can be cited without ranking, and rank without being cited.

What decides the citation is whether the page answers the sub-questions the engine asks on the buyer's behalf. A criteria-led list has that shape, with a field per question; an advert does not.

The chain has four links: the list is retrieved, then cited, then the author is mentioned, and sometimes recommended. Each link keeps only part of the one before it, and the self-ranking helps mainly at the third.

From list page to recommendation

1. List page published
2. Retrieved for a buyer prompt
3. Cited in the answer
4. Author brand mentioned
5. Author brand recommended

Should a B2B SaaS head of marketing write a category list?

Yes, write it, and write it as a buyer's guide rather than as an advert. The buyers are already asking: in G2's March 2026 survey of 1,076 B2B decision makers, 51% said they start software research in an AI chatbot more often than in Google.⁵ Semrush, which sells SEO and AI visibility software, surveyed 622 US B2B professionals in March and April 2026. Among the 519 who use AI at work, 92% said AI had shaped their vendor shortlist.⁶ That is the audience for the list: a buyer who asked an assistant for options before asking a peer, and who will check the answer against your site.

What the list should look like

The shape that worked in Scrunch's May to June 2026 data was a list of 4 to 10 brands, which got the author recommended more often than shorter or longer lists did.² Include the rivals a buyer would name unprompted, state the criteria and the date you applied them, and keep every price and feature current. A list that leaves out the obvious rival reads as an advert to the engine and to the reader.

The list is one of several pages the engines draw on for a software shortlist. Our guide to the seven pages a SaaS site must get right covers where it sits among pricing, integrations, security and alternatives pages. Our piece on 'X vs Y' comparison pages covers the two-brand version. The list earns its keep as part of an AI visibility programme in which every page says the same thing about you, rather than as a one-off.

Should you rank yourself first?

Only when your own criteria put you there, and the evidence says the gain is small either way. Ahrefs found self-ranked lists throughout ChatGPT's sources (period not stated), so engines do not refuse them.¹ Scrunch found the self-ranked list moved recommendations from about 4% to roughly 7% (May to June 2026).² Mentions rose from about 19% to about 39%, and a rival was named instead 24.3% of the time.²

The honest alternative is to rank by criteria and show the working. Put the comparison table before the prose, give every brand the same fields, and label the author's entry as the author's. If you top the table on the criteria a buyer cares about, say so. If you top it on one criterion, lead with that criterion and let the table rank the rest.

Our view: rank yourself where the criteria put you and publish the criteria. The lift from the top slot is mostly mentions. A number-one slot the table does not support is the claim an engine rebalances and a sceptical buyer discounts.

Google's rules apply to the list as much as to the answer

A category list is content, and Google's guidance on generative AI content now says it is critical to manually fact-check and review all AI-generated content before publishing.¹⁰ If a model drafts the list, a person checks every price, feature and integration against the vendor's own page before it goes live. Our note on Google's updated AI content guidance sets out the review record to keep.

Google's documentation on AI features says a page needs to be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode.¹¹ There are no extra technical requirements. Google's spam policies apply to all of its web search results, and scaled pages produced to manipulate rankings are named in them.¹³ A batch of thin category lists is the kind of page a spam update removes from the answer as well as from the ten blue links.

The buyer-side cost

There is a buyer-side cost as well. In G2's March 2026 survey, 69% of B2B decision makers said an AI chatbot had surfaced information that led them to choose a different vendor than they expected.¹⁴ A list that misstates a rival's pricing is one way to become that information, and to be corrected in the next answer. The rival's pricing page is the source the engine checks against, and the engine will side with it.

Our view: date the criteria, name the reviewer, and correct the list the week a vendor changes its pricing. The review a category list needs is the same one a product page needs.

How do you measure whether the list worked?

With a prompt set you run before the list goes live and again after, on the same engines, with the same method. Three runs per prompt per engine, on separate days, is the floor, because answers change from run to run and a single run is an anecdote.

1. Write twenty to thirty buyer prompts in the buyer's words, naming the category and the use case, never your brand. Rotate the countries you sell in, such as India and Germany.
2. Set the baseline: run each prompt on ChatGPT, Gemini, Microsoft Copilot, Google AI Overviews and Google AI Mode, three days apart. Record four things per answer: was your list cited, were you mentioned, were you recommended, and which rival was.
3. Publish the list with dated criteria and the rivals included, then leave it alone for at least four weeks.
4. Rerun the identical set and compare the four rates, prompt by prompt.
5. Track a control: a second category you did not touch, run the same way, so a general rise in mentions is not credited to the list.

What to read in the results

The two measures that matter are the recommendation rate and the rival named. Keep the raw answers, dated, so a change in a rival's position can be traced to the answer that showed it. A lift in citations and mentions with no movement in recommendations is the expected result, and it means the list is doing its job as a source and nothing more. The free score runs a smaller version of this check with our AI answer monitor; the audit runs the full set with confidence intervals.¹²

Our view: expect mentions to move before recommendations, and expect Gemini to move before Google's own answers do.² If recommendations have not moved after two reruns, stop polishing the list and fix the pages the engines recommend from, which are the ones a buyer would check next.