In B2B software, AI recommends from a pool of 3 to 5 brands per category

We ran 270 live AI prompt panels across three B2B software categories. Across all 270 panels, buyer-intent prompts produced only 12 unique brand recommendations. Within each category, the pool is even smaller. The AI recommendation list is not a ranking. It is a short list, and most brands are not on it.

The scope of the data

Each panel tested one brand across 10 buyer-intent prompts submitted to four AI engines: ChatGPT (with web search), Gemini (with grounding), Perplexity, and Claude (with web search). We measured who the AI recommended when the brand being tested was not recommended, tracking which competitors won the recommendation slot in each response.

The categories tested: project management software (173 brands), customer support software (58 brands), HR software (39 brands). 270 brands total. The question we were answering: across all of these buyer-intent queries, which brands actually appear as the AI's recommendation?

The recommendation pool by category

In HR software, across 39 brands tested and 117 recommendation instances, exactly three brands received any recommendations at all:

Brand	Recommendations received	Share of category
BambooHR	85	73%
Rippling	22	19%
Gusto	10	9%

Three brands. 39 were tested. These three accounted for 100% of the recommendation signals the AI engines produced for the HR software category.

Customer support software had a slightly wider pool, but the concentration is nearly identical:

Brand	Recommendations received	Share of category
Zendesk	88	51%
Freshdesk	66	38%
Intercom	12	7%
Others (2 brands)	8	5%

Five brands total. 58 were tested. Zendesk and Freshdesk together took 89% of all customer support recommendations.

Project management software has the largest brand set in the data (173 brands tested) and the smallest recommendation pool relative to the category size:

Brand	Recommendations received	Share of category
Asana	342	66%
Jira	85	16%
ClickUp	77	15%
Other (1 brand)	15	3%

Four brands. 173 tested. Asana alone received 66% of all project management software recommendations produced by AI engines across these panels.

This is not a long-tail market

One common assumption about AI recommendations is that, because AI engines have access to so much content, the recommendation set will be broader than traditional search. The assumption is that a niche brand with strong content could get recommended alongside the category leaders.

The data does not support this. In all three categories, buyer-intent AI recommendations are more concentrated than search rankings, not less. A buyer typing "best HR software for a 50-person company" into a search engine will see 10 organic results from many vendors. The same question to an AI assistant produces one or two named recommendations, drawn from a pool of three brands total.

This concentration appears to be a structural property of how AI engines answer buyer-intent questions. They retrieve sources, identify which brand is recommended in those sources, and surface the brand that appears as the answer most consistently across the retrieved content. The result is not a summary of the market. It is a weighted vote, and the voting power is concentrated in the sources the AI engine weights most heavily for each category.

What determines who is in the pool

The brands that receive AI recommendations are not simply the largest brands in the category. All three of the dominant HR software recommendations (BambooHR, Rippling, Gusto) are smaller than some brands in the test set that received zero recommendations. Recommendation pool membership appears to be driven by coverage and positioning in the sources that AI engines retrieve for buyer-intent queries: review aggregators, editorial comparison pages, and the handful of category-specific publications that AI engines weight most heavily.

A brand that ranks well on these sources for its category keywords will appear in the AI's retrieval set and therefore in its recommendations. A brand that relies primarily on its own domain content, even high-quality content, is less likely to be in the pool because AI engines responding to "what should I use" are retrieving third-party editorial sources, not the brand's own pages.

The implication for brands not in the pool

If your category has five brands in the AI recommendation pool and you are not one of them, improving your brand's AI mention rate will not fix the problem. Being mentioned more often in AI answers still leaves you in the majority group: brands that appear as context but not as recommendations. The gap that matters is between mention and recommendation, and the path from one to the other runs through the specific sources the AI engines use to answer buyer questions.

For most B2B software brands, the relevant question is not "what is our AI visibility score" but "are we on the short list, and if not, what would it take to get there." The short list in each category is short enough that you can identify it exactly, understand why those specific brands are on it, and determine whether a credible path to joining them exists.

Methodology note

270 B2B software brands measured in June 2026 using live, search-grounded AI endpoints: ChatGPT (OpenAI Responses API with web search), Gemini (with grounding), Perplexity (Sonar), and Claude (with web search). Each brand received 10 buyer-intent prompts specific to its category. Recommendation counting tracked instances where a competitor brand was named as the primary recommendation in response to a buyer-intent query. Branded queries, research-intent queries, and prompts that returned no named recommendation were excluded from the denominator. A recommendation counted once per response per competitor brand. Categories: project management software (173 brands), customer support software (58 brands), HR software (39 brands).