Do Comparison and Alternatives Pages Get You Into AI Recommendations?

The measured evidence is about being present in - or ranked highly within - other people's lists, not about publishing your own comparison pages. Two vendor studies report related but different associations. Peec AI links rank-tier exposure within frequently-cited third-party listicles to brand visibility. Position Digital links the share of category listicles a brand appears in to how often ChatGPT names it. Presence across lists and position within a list are not the same exposure. Both are observational, both say so, and both measure whether the brand was named - not whether the brand's own page was cited. Whether a vendor's own "X vs Y" or "alternatives to Y" page behaves like a third-party list has not been tested by anything we located.

That distinction is the whole page, because the obvious inference runs the wrong way. Evidence that appearing in someone else's ranked list is associated with being named does not license the conclusion that publishing your own ranked list will do the same thing. This is the comparison-content part of the vendor-shortlist pillar; the source-composition page owns the metric definitions this page uses without re-deriving.

Two labels mark evidence boundaries on this page. Documented means the statement it sits beside is directly supported by the linked platform documentation, quoted. Recommendation means a course of action that CoreAEX prescribes and no platform documents. Research findings are attributed in the prose with their sample, dates, engine and the outcome they measured, and are not tagged. Untagged text is ordinary explanation.

Two Different Things Get Called "Comparison Content"

Before any figure is usable, separate the two page types, because you control one of them and not the other.

• Third-party lists - "best project management tools," "top 10 CRMs," published by review sites, industry blogs, news publishers and independent editorial. You can be included in these; you do not write them, rank them or own them.
• Vendor-authored comparison content - your own "us vs them" page, your "alternatives to [competitor]" page, your use-case and integration pages. You write every word, and you are a participant in the comparison rather than an observer of it.

The published measurements cover the first category. One of them says so explicitly: the researchers state they focused exclusively on third-party listicles (such as independent review sites, industry blogs, and news publishers) that AI engines actually cite over and over again, rather than looking at every listicle on the internet. Nothing located measures the second category at all. A team that reads a finding about the first and acts on the second has substituted the thing it can control for the thing that was studied.

What Is Measured About Third-Party Lists

Two vendor studies measure different exposures on different outcomes, engines, samples and months, and report associations pointing the same broad way. They are worth reading together, and they are not replications of each other.

Rank within frequently-cited listicles, measured on brand mentions. Peec AI, an AI-visibility vendor, analysed nearly 200,000 AI responses producing over 5.7 million data points across three industries - B2B SaaS, emerging MarTech and US finance - on eight engines: ChatGPT, Claude Sonnet 4, Gemini, Google AI Mode, Google AI Overview, GPT-5 Search, Microsoft Copilot and Perplexity, between September 2025 and March 2026. Its outcome is defined plainly: Did the brand get mentioned in the AI's answer? Peec reports a positive association between higher rank-tier exposure in repeatedly-cited third-party listicles and brand visibility, and reports the visibility lift as largest in B2B SaaS among the three industries studied - other outcomes in the same report favour other industries, so the comparison is specific to the visibility metric, not a blanket ranking of the three.

No magnitude from that study appears on this page, and the reason is worth stating. The post reports percentage-point lifts by rank without saying what each is measured against - a lower rank, absence from the list, or some other comparison group. Its full specification sits in a linked companion working paper: Ehrlinspiel, Rudzki and Landwehr, "Cited-Listicle Rank-Tier Exposure, Author Type, and LLM Brand Visibility." On 31 August 2026 that paper's abstract page could not be loaded and its PDF endpoint returned 429 (rate-limited), so its reference category, model, controls and analysed sample could not be read. Until they can be, the direction is what the blog supports and the size is not usable - and neither is a comparison between it and any figure below.

The authors are unusually direct about what the design supports: the work is based on observational research, not a randomized experiment, and, having controlled for prompt, engine, date, brand-level differences and retrieval composition, state plainly that it is not causal proof. They also state that the findings apply to frequently retrieved listicles, not every listicle on the web - which narrows the population considerably, since the study deliberately selected lists that engines already cite repeatedly.

Listicle presence, measured on citations and on recommendations separately. A second, much smaller study by Position Digital, an SEO agency, ran 278 prompts across six B2B SaaS niches through ChatGPT using the third-party tool Ahrefs Brand Radar, capturing a single snapshot in July 2026 in the US and UK, covering 95 tracked brands and 3,508 citations from 2,845 unique pages. It separates two outcomes that this cluster keeps separate throughout: a citation is a source URL the engine references, and a recommendation is the brand being named in the answer text. Two figures matter here:

The 49.9% figure is the one that circulates, and it is three page types added together. Listicles alone account for 18.8% of citations in this sample. Quoting the combined number as a listicle finding inflates it by more than double.

The two correlations come from the same study on the same brands, so comparing them is legitimate in a way that comparing across studies is not: in this sample, the share of category listicles a brand appeared in tracked its recommendation frequency more closely than having its own site cited did. Both are bivariate associations in a single July snapshot on one engine, and the study says so itself: This is a small, directional study based on 278 prompts across six B2B SaaS niches in the US and UK. The data was captured as a single snapshot in July 2026. The findings reflect a specific segment at a specific point in time. They should not be treated as universal ranking factors or as rules that apply across the wider web.

Two more things belong on the record. The measurement depends on Ahrefs Brand Radar, a commercial tracking tool. And the publisher sells AI-search optimisation services, which does not disqualify the work but does mean the finding and the remedy come from the same place - the same disclosure this cluster applies to the review-platform evidence.

What Is Known About Your Own Comparison Pages

We located no study that measures any outcome for vendor-authored comparison pages as a category. That is the honest answer, and it is worth stating as precisely as an absence can be stated - including what was searched for.

We searched for research measuring whether vendor-authored comparison content - "us vs them" pages, "alternatives to [competitor]" pages, use-case pages - is cited, mentioned or recommended at a different rate from other content a vendor publishes. We located no study that isolates vendor-authored comparison pages as a category and measures any outcome for them. The Peec study explicitly excluded them by design. The Position Digital study classifies citations by page type across the whole sample rather than by who wrote the page, so its listicle share includes third-party lists and cannot be read as a finding about vendor-owned comparison pages.

This is an absence of evidence, not a finding of no effect. Your comparison pages may work well, may work poorly, or may behave exactly like third-party lists. Nobody has measured it, and the two possibilities are not symmetrical in the way people assume: a third-party list is a source an engine may treat as independent of the vendors in it, while your own comparison page is written by a participant. Whether that difference matters to a retrieval system is precisely the untested question.

One adjacent result is worth carrying as a caution rather than an answer. In a preprint studying 112 recently-launched products through two model API endpoints in December 2025, a composite score of on-page optimisation for AI visibility showed no statistically significant correlation with discovery on either endpoint. That is a detected null on a regex-based measure in a narrow sample - it does not show that page-level work is ineffective, and it is not a finding about comparison pages specifically. What it does supply is a reason not to assume that adopting a page format is itself the mechanism; the full statistical detail behind this finding is set out in why your SaaS brand is missing from AI recommendations.

Being Cited and Being Named Are Different Goals

In one small ChatGPT-only sample, two thirds of brand recommendations occurred without the brand's own site being cited at all. That figure does more to clarify what a comparison page is for than any other on this page.

In the Position Digital sample, 66% of brand recommendations happen without ChatGPT citing that brand's website at all. One engine, one July snapshot, 278 prompts, six niches, US and UK - a narrow finding, and a clarifying one. It means the goal a comparison page is usually built for - getting your own URL into the sources list - is not the same as the outcome a team is usually after, which is being named as a candidate.

Those two goals point at different work. Being cited depends on your page being retrievable, relevant to the question asked, and useful enough to reference. Being named can happen through any source the engine consults, including lists you did not write and pages you do not own. The source-composition evidence points the same way: in one small brand-level benchmark, external sources made up the large majority of the cited-source mix for SaaS brands.

The practical consequence is uncomfortable for a content plan. The lever with measured evidence behind it is being included, and ranked well, in lists other people publish - which is editorial and relationship work, not page production. The lever a content team can execute unilaterally is the one nobody has measured.

Figures This Page Does Not Use, and Why

Three widely-shared claims about page types are absent here deliberately. They are the ones a reader searching this topic is most likely to meet.

"AI engines cite listicles at five times the rate of blog posts." This comes from a press release rather than a published study. Read in full on 31 August 2026, its stated basis is that a single ranked listicle page accumulated 294 citations from AI search engines while conventional blog posts covering adjacent topics in the same vertical collected between 15 and 91 citations each over seven consecutive days. The numerator is one page. No dates, no geography, four of the eight claimed engines unnamed, no methodology or data published, and no limitations stated. It also presents a raw count as a "rate" - 294 citations is a total, not a rate, because no denominator of queries or opportunities is given. The direction may well be right; the evidence offered cannot establish it.

Page-type multipliers with inconsistent attribution. One frequently-cited page reporting which page types earn citations mixes clearly-sourced external studies with figures credited only to internal research, carrying no sample, window, geography or methodology link. On the same page, a headline multiplier appears in a table with no source given at all, while the same figure is correctly attributed to the external study elsewhere in the page's body text - a reader scanning the table alone cannot tell which source produced it. An ambiguous number is not a weak number; it is one a reader has to hunt to check.

Citation counts drawn from prompts that select for lists. Another analysis counts over 10,000 citations across 57 queries chosen because they contain "tools," "software" or "alternatives to." Prompts asking for tools and alternatives will disproportionately return list-format pages, so the resulting page-type distribution is substantially a property of the prompt set. On a page about whether list formats earn citations, that is the specific artefact to avoid, and while the article states a February 2026 publish date and its query count, it gives no collection-period dates for the underlying citations, no geography and no significance testing against which the finding could be checked.

What Follows for a SaaS Team

The evidence supports a reallocation, not a content sprint. Three things follow, and they are ordered by how much measurement stands behind them.

Treat third-party lists as the prioritised opportunity set, not as a proven lever. The measured association concerns presence across lists - and, in Peec's study, rank-tier exposure within them. Both studies are observational, so neither shows that gaining or improving a placement will cause your recommendation frequency to rise. What they justify is where to look first. The work that follows is unglamorous: knowing which lists in your category get cited repeatedly, making sure your product is in them, and making sure what they say about you is accurate and current. Then measure any change with a repeated panel, and do not forecast a lift from the published coefficients. Recommendation

Publish your own comparison and alternatives pages for the reasons that have always justified them. Buyers in a comparison mindset look for them; sales teams use them; they answer questions your product pages avoid. Those are sufficient reasons. What you cannot currently claim is a measured effect on AI recommendations, because nobody has measured one - so build them to be accurate, specific and genuinely useful about the competitor as well as about you, and do not budget them against a citation forecast. Recommendation

Separate "get cited" from "get named" in whatever you measure. If most recommendations in your category happen without your site being cited, a programme optimised only for citations is measuring the narrower outcome. The measurement page covers the panel design that records these as separate coded columns rather than merging them, and the variation page covers how many runs it takes before any before-and-after reading means anything. Recommendation

One thing not to do: do not treat a page format as a mechanism. No provider documentation located names a preferred page type, and Google's guidance for its AI features states that eligibility rests on ordinary indexing and snippet eligibility, with no additional technical requirements. Documented Formatting a page as a numbered list because lists appear in citation counts is copying the surface of a correlation.

What Is Not Known

The central question of this page is untested. No located study isolates vendor-authored comparison, alternatives or use-case pages and measures citations, mentions or recommendations for them. Every figure here describes third-party lists or page types classified without regard to who wrote them.

No study located varies a page and measures what changes. Both studies on this page are observational and both say so. Nobody has published a design in which a comparison page is added, removed or rewritten and the effect on an AI answer is observed against a control, for B2B software.

The magnitude of the rank-tier finding has not been checked here. Peec's post reports lifts by rank without stating a comparison group, and the specification sits in a linked working paper whose full text could not be retrieved on the review date. What the blog supports is a reported positive association on an observational design. The size, its reference category and its model remain unread, and no figure from that study is printed on this page.

Both studies come from publishers selling services in this field, one measuring through its own platform and the other through a commercial tracking tool. Neither is independent measurement. The two teams appear institutionally separate, but the studies are not replications of each other: they use different exposures, outcomes, engines, samples, tools and months. Their results therefore cannot be pooled, and agreement in broad direction is weaker evidence than two studies measuring the same construct would be, not stronger. Their figures are not pooled anywhere on this page and should not be pooled anywhere else.

Sources

Sources: Platform documentation, quoted as read on August 31, 2026: Google Search Central, AI features and your website (last updated 2025-12-10) and Google's Guide to Optimizing for Generative AI Features on Google Search (last updated 2026-07-10) - neither names a preferred page type. Vendor and practitioner research, each named as such: Peec AI, "The Listicle Rank Effect" (published May 14, 2026; nearly 200,000 AI responses and over 5.7 million data points across B2B SaaS, emerging MarTech and US finance; eight engines - ChatGPT, Claude Sonnet 4, Gemini, Google AI Mode, Google AI Overview, GPT-5 Search, Microsoft Copilot and Perplexity - between September 2025 and March 2026; geography not stated; brand and listicle counts not stated; the outcome is brand mention, defined as "Did the brand get mentioned in the AI's answer?"; the exposure is rank-tier within those listicles; the study covers only third-party listicles that engines cite repeatedly, and states that it is "based on observational research, not a randomized experiment." Its full specification is in a linked companion working paper - Jan Ehrlinspiel, Tomek Rudzki and Malte Landwehr, "Cited-Listicle Rank-Tier Exposure, Author Type, and LLM Brand Visibility: A Two-Part Model of Selection and Prominence in Generative Engine Responses," SSRN working paper 6753841. That full text could not be retrieved on August 31, 2026 (abstract page inaccessible, PDF endpoint returned 429), so its reference category, model, controls and analysed sample are unread and no magnitude from this study is used on this page); and Position Digital, "Top ChatGPT Ranking Factors in B2B SaaS" (published August 6, 2026; 278 prompts across six B2B SaaS niches, 95 tracked brands, 3,508 citations from 2,845 unique pages; ChatGPT only, run through Ahrefs Brand Radar; a single snapshot in July 2026, US and UK; citations and recommendations are separated but not formally defined; the authors state the work is "a small, directional study" whose findings "should not be treated as universal ranking factors"; publisher sells AI-search optimisation services). Preprint, not peer-reviewed: Amit Prakash Sharma, "The Discovery Gap" (arXiv 2601.00912v1; 112 startups, 2,240 queries via API against gpt-4o-mini and sonar with web search, December 15-20, 2025; the composite on-page optimisation score showed no statistically significant correlation with discovery on either endpoint, and the author states that the scoring was regex-based - the full statistical detail is set out in why your SaaS brand is missing from AI recommendations). Sources described in "Figures this page does not use" were read in full on August 31, 2026 and are not relied on for any claim. No measurement cited here establishes that any page format causes a vendor to be cited, named or recommended, and figures from different studies are not pooled or compared numerically anywhere on this page.

---