More software buyers now start with a question to an AI assistant instead of a search box: "what's the best help desk for a small team?", "Jira alternatives for a startup?". So we asked. From 7 to 13 October 2026 we put the same buyer questions to ChatGPT, Claude, Google AI Mode and Perplexity every day, across 12 software categories, and recorded every product each answer recommended. That came to 3,670 answers.
The short version: the four assistants disagree far more than most marketers assume, and they get their information from very different corners of the web.
1. In 4 of 12 categories, each AI has a different favourite
Ask for the best customer support platform and every assistant's most-recommended product is different. ChatGPT names Intercom most often, Claude Freshdesk, Google AI Mode Zendesk and Perplexity Help Scout.
| Brand | ChatGPT | Claude | Google AI Mode | Perplexity |
|---|---|---|---|---|
| Intercom | 79% | 60% | 50% | 40% |
| Freshdesk | 61% | 80% | 51% | 50% |
| Zendesk | 75% | 72% | 67% | 60% |
| Help Scout | 75% | 74% | 50% | 71% |
Project management splits three ways: Linear on ChatGPT (92% of answers), Jira on Claude and Perplexity, and ClickUp on Google AI Mode. Website builders (Webflow vs Wix) and form builders (Tally vs Jotform) split two ways.
2. Being recommended by one AI tells you little about the others
Some of the biggest gaps we saw for products in their category's top eight:
- Notion appears in 68% of ChatGPT's project-management answers and 8% of Perplexity's.
- Dashlane appears in 80% of Claude's password-manager answers and 22% of Perplexity's.
- Cal.com appears in 96% of Claude's scheduling answers and 45% of Google AI Mode's.
- Attio appears in 73% of ChatGPT's CRM answers and 37% of Google AI Mode's.
- WordPress appears in 78% of ChatGPT's website-builder answers and 32% of Perplexity's.
If you check your brand in ChatGPT and stop there, you are seeing a quarter of the picture.
3. Where they agree, it's close to unanimous
In the other eight categories all four assistants share a favourite, and in seven of them it shows up in 90% or more of answers: Notion for note-taking (99% of answers), Figma for UI design (97%), Bitwarden for password managers (96%), Calendly for scheduling (94%), beehiiv for newsletters (92%), HubSpot for CRM (90%) and Loom for screen recording (90%). For buyers asking these questions, AI has already picked a default.
4. Challengers are beating the incumbents
- Web analytics: Plausible is recommended in 82% of answers, Google Analytics 4 in 43%. Matomo, Umami and Fathom all rank above GA4 too.
- Newsletters: beehiiv 92%, Mailchimp 32%.
- Password managers: Bitwarden is named more often than 1Password (96% vs 88%), though 1Password is put first more often (232 first places to 122).
These questions don't name a brand, so nothing in the prompt favours the incumbent. The answers reflect what the assistants have read about each category, and in several categories that favours newer tools over the household names.
5. They read completely different parts of the web
We also logged the sources cited in 3,193 of these answers: 1,654 distinct websites in all.
- Google AI Mode leans on YouTube and Reddit. It cited youtube.com in 28% of its answers and reddit.com in 24%. ChatGPT and Claude never cited either.
- ChatGPT leans on small comparison sites. Its most-cited sources were dupple.com (16%), toolradar.com (9%), itechguides.com (8%), stackfyi.com (7%).
- Claude cited costbench.com (9%), toolradar.com (9%), dupple.com (9%) most often.
- Perplexity leans on zapier.com (16% of answers) and vendors' own sites.
- Review sites barely appear. g2.com was cited in 2% of Google AI Mode answers and 1% of ChatGPT's; capterra.com in 2% of Claude's at most.
What this means if you sell software
- Check each assistant separately. A strong showing on one says little about the other three.
- Google AI Mode is won on Reddit and YouTube, where a real discussion or a walkthrough video is more likely to be quoted than your own site.
- ChatGPT and Claude are won on comparison pages. Being included, accurately, in the "best X for Y" articles they read matters more than a review-site profile.
How we measured, and the caveats
- Each category has 10 buyer questions that never name a brand, asked once a day to each assistant through its normal web interface (ChatGPT, Claude and Perplexity on free accounts, Google AI Mode signed out), from the US, with web search switched on. Claude still answers many questions without citing any source.
- A product counts as recommended when the answer presents it as an option. Share means the percentage of that assistant's answers in the category that recommend it.
- ChatGPT contributed fewer answers this week (613 of the 3,193 in the citation sample) because a ChatGPT interface change interrupted our collection for two days. Its shares are still computed over its own answers.
- AI answers vary from run to run. A week of daily answers smooths that out, but treat single-digit differences as noise.
The live version of this data, updated daily, is on our AI rankings pages, with a page for every brand mentioned above.
About us
NeverDrafts is an AI search visibility platform, so we have an interest here. It runs this same measurement for one brand's own buyer questions, then writes and publishes the content aimed at the questions that brand is losing, and re-measures. If you want to see where your brand stands today, the AI visibility audit is free and needs no card.