We asked five engines for the best CRM, 600 times. Here's who got named.

We asked five answer engines the same question buyers ask every day — some version of "what is the best CRM for a growing sales team" — and we asked it 600 times. Across ChatGPT, Claude, Gemini, Perplexity and Google's AI Overviews, in the United Kingdom, Ireland and the United States, over a fortnight in May. Then we recorded, for every answer, which brands were named, in what order, and which sources the engine reached for to get there.

The point was not to crown a winner. It was to read the water: to see who the engines reach for, how much they disagree, and where a brand's share of voice actually sits when nobody is performing for an analyst. The findings below are drawn from synthetic-but-representative runs — the kind of result our instrument returns on a real category.

The shortlist is shorter than the market

Across 600 answers, 31 distinct CRM brands were named at least once. But the distribution was brutally top-heavy. The four most-named brands accounted for 71% of all mentions. Eleven brands were named exactly once.

• Salesforce — Share of voice 26% · First-mention rate 41%
• HubSpot — Share of voice 24% · First-mention rate 38%
• Pipedrive — Share of voice 12% · First-mention rate 9%
• Zoho CRM — Share of voice 9% · First-mention rate 4%
• monday CRM — Share of voice 6% · First-mention rate 2%
• Everyone else (26 brands) — Share of voice 23% · First-mention rate 6%

The headline: two incumbents took half the room. But the more useful number is the gap between share of voice and first-mention rate. HubSpot was named almost as often as Salesforce, yet led the answer slightly less often — the engines reach for it constantly but tend to position it second. Pipedrive's story is the inverse of the long tail's: a respectable 12% share, but it is rarely the name an engine opens with.

The engines do not agree with each other

The single most important finding for any brand: there is no such thing as "the AI answer". There are five, and they diverge sharply.

• ChatGPT was the most incumbent-friendly, giving Salesforce and HubSpot a combined 58% share and rarely venturing past the top five.
• Claude was the most hedged — it named more brands per answer (3.8 on average versus ChatGPT's 2.9) and was the most likely to add qualifying language about company size and budget.
• Gemini leaned hardest on Google's own surrounding context and surfaced more region-specific names, including UK-focused tools that the US-trained pattern of the others ignored.
• Perplexity was the most source-transparent and the most volatile: its shortlist changed most between identical prompts run a week apart, tracking whichever comparison article had been published most recently.
• AI Overviews were the most conservative of all, frequently naming just two brands and citing a single dominant source.

A brand sitting at 12% overall share can be at 19% inside Gemini and 4% inside ChatGPT. Measuring one engine and generalising is the central mistake — the variance between engines is larger than the variance between most mid-tier brands.

If you optimise for the average answer, you optimise for an answer no engine actually gives.

The sources that moved the models

We logged 1,180 source citations across the runs. They concentrated almost as tightly as the brands did. Five domains accounted for 64% of all citations.

• g2.com — Share of citations 22% · Notable for Drove first-mention order more than any other source
• Reddit — Share of citations 16% · Notable for Heavily weighted by Perplexity; near-absent in AI Overviews
• Vendor's own site — Share of citations 13% · Notable for The clearest "what we do, who for" pages won here
• Wikipedia — Share of citations 8% · Notable for Set category membership — who counts as a CRM at all
• TechRadar / comparison media — Share of citations 5% · Notable for Recency-sensitive; moved Perplexity most

The pattern is consistent with what we see across categories. Third-party review aggregators set the order, community discussion adds or removes the challengers, and the vendor's own pages decide whether the engine can describe the brand clearly enough to name it with confidence. The brands that punched above their market share — Pipedrive especially — almost always had unusually clean, specific product pages and a strong, recent G2 presence.

Geography is a live variable

The same prompt, asked from Dublin, named a different shortlist than the same prompt asked from Chicago. UK and Irish runs surfaced region-aware tools and were noticeably more likely to mention pricing in pounds and data-residency considerations. US runs defaulted to the global incumbents and rarely localised at all. For any brand selling across borders, a single-country read is a partial read.

What a brand should take from this

Three things, in order.

1. Presence is concentrated, so the prize for entering the shortlist is large. With four brands holding 71% of mentions, moving from the long tail into the top eight is a step-change in visibility, not a marginal gain.
2. The engine you are weak in is the one to measure first. Because the engines disagree so much, your lowest score is usually your biggest, cheapest opportunity — not your average.
3. The levers are the sources. A strong, recent G2 profile, clear vendor pages, and presence in the community threads the engines actually read will move the answer faster than anything you can say about yourself in isolation.

None of this is visible in a traffic dashboard. It is visible only by asking the engines the same questions buyers ask, over and over, and writing down what they say. That is the read. The water is moving — the only question is whether you are measuring it.