Kubnal Bridge
Back to Blog

How Do AI Engines Decide Which SaaS Vendors to Recommend?

Kubnal Bridge Editorial TeamSeptember 30, 202610 min readB2B SaaS
Title card reading How Do AI Engines Decide Which SaaS Vendors to Recommend?, from kubnalbridge.com

AI engines decide which SaaS vendors to recommend by searching first. They read a small set of pages, then name a shortlist of five or six vendors. Naming and citing are separate steps. The head of the shortlist is largely shared across engines, while the pages each engine cites are its own.

If you're looking for a list of ChatGPT ranking factors, there isn't one. OpenAI publishes no such list. What exists are patterns you can observe and correlations you can measure. We measured those patterns in our audit of 25 software-buying prompts across ChatGPT, Perplexity, and Google AI Overviews.

Key Takeaways

  • There's no published ChatGPT ranking-factor list. The engines search, read a handful of pages, and name a shortlist of five or six vendors.
  • Naming and citing are separate steps. On each engine we tested, most named vendors were recommended without their own site being cited.
  • On ChatGPT, the top of the shortlist got cited and the tail mostly didn't: 64% of first-named vendors versus 24% named fourth or later.
  • The engines agreed on the head of the shortlist and split on the rest.
  • A few vendors took most mentions in each category, which suggests a consensus across the web rather than any single page.

What Do AI Engines Do Before They Name a Vendor?

AI engines search the web before they name a vendor. Across our 25 prompts, ChatGPT ran 90 searches and retrieved 1,056 results, then cited about four sources per answer. Perplexity cited about 20 sources per answer and Google AI Overviews about seven. On ChatGPT, the cited sources are a small slice of what it read.

Our 25 prompts covered five software categories: CRM, project management, help desk, HR and payroll, and developer monitoring. Each category was asked in five styles: a plain "best X", company size, integration, price, and "alternatives to" the market leader. We ran ChatGPT logged out from the US with search on, Perplexity through its Sonar API, and AI Overviews on US desktop Google. The headline results are in our pillar guide to GEO for B2B SaaS.

The domains ChatGPT retrieved most often were vendors' own sites and help centers, such as freshworks.com, zendesk.com, and docs.helpscout.com. Price was a frequent theme: 56 of its 90 searches, or 62%, contained a price word. So your pricing page is part of what ChatGPT reads. Our guide to SaaS pricing pages and AI recommendations covers it.

The shortlists that came out were short. Answers named 5.3 to 5.6 vendors on average, depending on the engine, with a minimum of 3 and a maximum of 12. A buyer who asks for the best CRM sees about five names. If yours isn't one of them, that answer never puts you in front of the buyer.

Do you need to rank on Google to be named in an AI Overview? On our prompts, most named vendors didn't. Of the 112 vendors the AI Overviews named, only 21, or 18.8%, had their own domain anywhere in that page's organic results. The rest reached the answer without a ranking page of their own.

Google's documentation suggests one reason. AI Overviews and AI Mode "may use a 'query fan-out' technique," which Google describes as "issuing multiple related searches across subtopics and data sources." It doesn't say how often. An answer built from several searches can name vendors that one results page never shows.

Why Are Most Recommended Vendors Never Cited?

Most recommended vendors go uncited because naming a vendor and citing a source are separate steps. In our audit, the vendor's own site was cited in the same answer for 41.1% of named vendors on ChatGPT. The figure was 42.0% on Perplexity and 20.5% on Google AI Overviews. The rest were named on the strength of someone else's pages.

This creates a puzzle on ChatGPT. Of its 103 citations, 65.0% were on vendor domains, so vendor pages dominate what it cites. Yet 58.9% of the vendors it named had no own-site citation in that answer. How can both be true?

Part of the answer is arithmetic. ChatGPT cited 4.1 sources per answer on average while naming 5.6 vendors, so it couldn't give every vendor its own page. The rest is where those citations went. Only 62 of its 103 citations were on named vendors' sites, and they covered 58 vendors, 55 of them with exactly one page. The citations run out well before the shortlist does.

The other engines lean on different evidence. Perplexity reads lists: 81.9% of its citations were "best," "top," or "alternatives" lists. Google AI Overviews read forums and video. They cited Reddit in 12 of 21 answers and YouTube in 6. Review sites such as G2 and Capterra showed up in 14 of 25 Perplexity answers, but only 2 on each of the other engines.

What does this mean for your product? A recommendation doesn't require your page to be cited, and a cited page doesn't guarantee a recommendation. The two steps need separate work. If ChatGPT names your competitors and not you, our diagnostic on why ChatGPT isn't recommending your SaaS product walks through the usual causes.

Does Your Place in the Shortlist Change Whether You're Cited?

On ChatGPT, a vendor's place in the shortlist tracked whether its own site was cited. The site was cited for 64.0% of first-named vendors, 60.0% of second-named, and 44.0% of third-named. For vendors named fourth or later, it was 24.2%. Perplexity and Google AI Overviews showed no such gradient.

We ordered vendors by where each was first mentioned in the answer text. That usually matches a numbered list, though not always, since an intro sentence can name a vendor before the list begins. On Perplexity the rates ran 44.0%, 52.0%, 32.0%, and 41.3% by position. On AI Overviews they ran 19.0%, 19.0%, 9.5%, and 26.5%.

ChatGPT: named vendors whose own site was cited, by position in the answer (Kubnal Bridge SaaS citation audit) ChatGPT: named vendors whose own site was cited By position in the answer 1st named 64% (16 of 25) 2nd named 60% (15 of 25) 3rd named 44% (11 of 25) 4th or later 24% (16 of 66) Source: Kubnal Bridge SaaS citation audit, 25 software-buying prompts. Order by first mention in the answer.
On ChatGPT, the higher a vendor sat in the answer, the more often its own site was cited.

Our reading: ChatGPT backs its top picks with their own pages and lets the tail ride on other sources. Pricing pages were its most-cited vendor page type, so we suspect many of those backing pages were pricing pages. Vendors lower down still made the list, probably on the strength of the independent lists and comparisons it also cited.

Why would position and citation move together? One possibility: the engine puts the vendors it has the most evidence for at the top, and its searches supply that evidence. Another is that the vendors already at the top of the category are simply the ones with the most searchable pages. Our data can't separate those. It's a correlation from one audit, and we read it as one. Either way, a challenger's target is the same: move up the list, and give ChatGPT a page worth citing when you get there.

Position Digital's ChatGPT study found a similar gap. In its data, 66% of brand recommendations happen without ChatGPT citing that brand's website at all. It also found that the share of category listicles a brand appears in correlates with how often ChatGPT names it, at r = +0.64. That's a correlation with mention frequency, not rank. The study covered 278 prompts in six B2B SaaS niches, in the US and UK, from one July 2026 snapshot of ChatGPT alone.

In our ChatGPT answers, most uncited recommendations sat in the tail. If you're named fourth or later, expect the answer to rest on someone else's page.

Do ChatGPT, Perplexity, and Google Agree on the Shortlist?

ChatGPT, Perplexity, and Google largely agree on the head of the shortlist and split on the rest. On the 21 prompts all three answered, there were 189 vendor-prompt pairs. All three engines named 57 of them, and only one engine named 90. The first-named vendor was usually one all three engines named.

A vendor-prompt pair is one vendor named for one prompt, by any engine. The first-named vendor was an all-three vendor in 16 of 21 ChatGPT answers and 17 of 21 Perplexity answers. In AI Overviews, it was 15 of 21. So when an engine opens its list, it usually opens with a vendor the other two also named.

How many engines named each vendor, per prompt (Kubnal Bridge SaaS citation audit) How many engines named each vendor Vendor-prompt pairs on the 21 prompts all three engines answered All three engines 57 (30%) Two engines 42 (22%) One engine only 90 (48%) Source: Kubnal Bridge SaaS citation audit, 25 software-buying prompts. Each vendor counted once per prompt.
Nearly half of the vendors named on a prompt came from just one engine.

This matches the overlap figures in our pillar guide. Named vendors overlapped by 40.8% to 48.5% on average between any two engines, while cited domains overlapped by only 3.0% to 9.0%. The engines reach similar names from mostly different pages.

On ChatGPT, consensus and citation went together. ChatGPT cited the vendor's own site for 50.9% of the vendors all three engines named. For vendors named by two engines, the rate was 37.1%, and for vendors only ChatGPT named, 26.9%. The more engines named a vendor, the more likely ChatGPT was to cite its site. Perplexity and AI Overviews showed no such pattern.

The shared head seems to reflect a consensus the engines pick up from many pages across the web. The single-engine tail is where each engine's own sources decide who gets in. For your team, that means working on both: the wider consensus and the pages each engine reads.

Why Do the Same Few Vendors Keep Winning?

The same few vendors keep winning, we think, because the head of the shortlist reflects what the wider web already says. In CRM, the five most-named vendors took 71.2% of all vendor mentions, and HubSpot was named in 14 of 15 CRM answers. In help desk, Freshworks appeared in all 14 answers across the three engines.

Every category was concentrated, though some more than others. The five most-named vendors took 62.8% of mentions in help desk and 60.9% in HR and payroll. In developer monitoring it was 51.6%, and in project management 50.6%. Across all three engines we counted 91 distinct vendors: 57 on ChatGPT, 72 on Perplexity, and 50 on AI Overviews. That's a long list, with a short head.

In our audit, a buyer asking any of the three engines for a CRM almost always saw HubSpot. The names that differed sat further down the list. That's a lot of shelf space for the leaders. A newer product can be good and still feel invisible, because it's competing for the few slots the leaders leave open.

Does that mean the engines favor big brands? We'd put it differently. What our data shows is a small group of vendors that every engine keeps naming. That fits a web where incumbents have gathered mentions in lists, reviews, and forums for years. The advantage is real, but it seems to run through what other pages say about you. And that's something a challenger can work on.

A 2025 Ahrefs study of 75,000 brands fits that reading for Google. Of the factors it tested, branded web mentions tracked brand mentions in AI Overviews most closely, with a Spearman correlation of 0.664. Backlinks correlated at 0.218. The study covers Google AI Overviews only, not ChatGPT. And Ahrefs itself cautions that correlation isn't causation.

Incumbents win the head, but qualifiers open the list. Against the same engine's plain "best X" answer, 38.5% of vendors named on company-size prompts were new. So were 47.0% on integration prompts, 38.2% on price prompts, and 53.5% on alternatives prompts. Add a condition, and more than a third of the shortlist changes.

Where a Challenger SaaS Can Break In

A challenger can't buy a place in the consensus. It can build the evidence each engine reads and target the qualifier prompts where shortlists open up. On our prompts, 38.2% to 53.5% of the vendors named on qualifier prompts were new compared with the plain answer.

Match the work to the engine:

  • Lists written for your qualifiers. Perplexity's citations were mostly lists, so target the ones for your buyer's size, integrations, or budget. Our guide covers getting your SaaS into "best tools" lists.
  • Complete review profiles. Review sites appeared in 14 of 25 Perplexity answers and rarely anywhere else. Our guide asks whether G2 and Capterra reviews decide AI recommendations.
  • A genuine Reddit presence. AI Overviews cited Reddit in 12 of 21 answers, and ChatGPT in none. Our guide on whether ChatGPT uses Reddit to recommend SaaS tools explains the split.
  • Clear pricing and product pages. ChatGPT backs its top picks with vendors' own pages, and pricing pages were its most-cited type of vendor page.

Our audit is a single snapshot: one answer per engine per prompt, from one logged-out US setup, across 25 prompts. We haven't measured how much answers change between runs. The links between position, consensus, and citation are correlations from that snapshot, so read them as patterns to test rather than rules.

Frequently Asked Questions

How does ChatGPT decide which brands to recommend?

ChatGPT searches the web before it answers. Across our 25 software-buying prompts, it ran 90 searches and retrieved 1,056 results. It then named 5.6 vendors per answer on average and cited about four sources. Vendors near the top of its list were the most likely to have their own site cited.

Is there a list of ChatGPT ranking factors?

No. OpenAI publishes no ranking-factor list, so what exists are observed patterns and correlations. According to Position Digital's study, 66% of brand recommendations happened without ChatGPT citing that brand's website. Our audit found a similar gap: 58.9% of the vendors ChatGPT named had no own-site citation.

Does ranking on Google mean AI will recommend you?

Not necessarily. In our audit, only 18.8% of the vendors named in Google AI Overviews had their own domain in that page's organic results. According to Google's documentation, its AI features may run several related searches for one answer. So named vendors can come from beyond the page you see.

Does ChatGPT favor big brands?

In our audit, a few vendors took most mentions in each category, across all three engines. In CRM, the five most-named took 71.2%, and HubSpot appeared in 14 of 15 answers. But qualifiers open the list. When a prompt added size, an integration, price, or a rival, 38.2% to 53.5% of named vendors were new.

Do all AI engines recommend the same vendors?

Partly. On the 21 prompts all three engines answered, 57 of 189 vendor-prompt pairs were named by all three, and 90 by only one. Named vendors overlapped by 40.8% to 48.5% between engines, but cited domains by only 3.0% to 9.0%.

Conclusion

AI engines decide which SaaS vendors to recommend in two steps. The head of the shortlist is a consensus the engines largely share, while the citations behind each answer differ by engine. Work on both: your presence in the pages that shape the consensus, and owned pages each engine can cite.

Agencies such as Kubnal Bridge specialize in mapping which pages each engine reads for your category and where your product falls in the shortlist. That mapping is a sensible first step if you're unsure where you stand.

About the Author

Kubnal Bridge Editorial Team. The Kubnal Bridge editorial team researches and writes about generative engine optimization and AI search visibility across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude.

Sources

Back to all articles