AI Summary
We asked ChatGPT, Google Gemini, Perplexity, and Google’s AI Overview the same simple buying question, then did it again with a second one. Four engines, four different shortlists. Only two or three brands showed up on every list. Everything below that was a coin flip, and one engine dropped the category’s biggest name entirely.
This used to be a footnote. It is now the central problem, because your buyers have spread across engines. In June, ChatGPT’s share of AI assistants slipped below 50 percent for the first time, to 46.4 percent, with Gemini at 27.7 percent and Claude at 10.3 percent, according to Sensor Tower data reported by TechCrunch. In the last two weeks alone, Anthropic shipped Claude Sonnet 5, and OpenAI launched ChatGPT Work and made GPT-5.6 the default model inside Microsoft 365 Copilot.
The referral data confirms it. A July study of 6.77 million sessions found ChatGPT still drives about 92 percent of standalone AI referral traffic, while Claude quietly overtook Perplexity for second place. A separate brand-averaged panel study put ChatGPT much lower, at 63 percent, with Claude at 18.5 percent. The two disagree on the exact split. They agree on the direction: your buyers are spreading across engines, and each engine answers differently. That is the backdrop for our state of AI search tracking, and it is why a single-engine strategy misses a growing share of your buyers.
How we tested
We picked two ordinary buying questions a small business would actually ask, “what are the best CRM platforms” and “what are the best email marketing platforms,” and put the exact same prompt into all four engines from a United States session in July 2026. We recorded which brands each one named in its top five and in what order. Here is how often each brand appeared across the four.
Finding 1: the top is safe, the tail is a lottery
For the best CRM question, the top of every list looked familiar. HubSpot, Pipedrive, and Zoho showed up in all four engines. That is the safe core, the set of names so established that no engine leaves them out. It matches what we see in our hands-on AI visibility tool testing: category giants are hard to dislodge.
| Engine | Its top CRM picks, in order |
|---|---|
| ChatGPT | HubSpot, Pipedrive, Zoho, Salesforce, Monday |
| Google Gemini | HubSpot, Pipedrive, Zoho, Monday, Freshsales |
| Perplexity | HubSpot, Zoho, Pipedrive, Salesforce, Bigin |
| Google AI Overview | HubSpot, Pipedrive, Salesforce, Zoho, Salesflare, Capsule |
Look past the top three and the agreement falls apart. Gemini left Salesforce off its list completely, which is the world’s largest CRM by revenue. Perplexity was the only engine to name Bigin. Google’s AI Overview was the only one to surface Salesflare and Capsule. Monday made two lists, Freshsales made one. If your brand is not one of the three untouchable names, whether an AI recommends you is decided engine by engine.
Finding 2: Google’s AI Overview diverged from the other three
The email marketing question split the engines even harder. The three chat assistants, ChatGPT, Gemini, and Perplexity, clustered around MailerLite, Brevo, Mailchimp, and ActiveCampaign. Google’s AI Overview went somewhere else entirely. It was the only engine to recommend Klaviyo and ConvertKit, and the only one to leave MailerLite and ActiveCampaign off the list.
| Engine | Its top email picks, in order |
|---|---|
| ChatGPT | HubSpot, Mailchimp, Brevo, MailerLite, ActiveCampaign |
| Google Gemini | MailerLite, Brevo, Mailchimp, Omnisend, ActiveCampaign |
| Perplexity | MailerLite, Brevo, Mailchimp, ActiveCampaign, Moosend |
| Google AI Overview | Mailchimp, Klaviyo, Brevo, ConvertKit |
Three different brands led the four lists. Google’s AI Overview led with Mailchimp, ChatGPT led with HubSpot, and both Gemini and Perplexity led with MailerLite. Only Mailchimp and Brevo made all four lists. A brand like Klaviyo, which dominates e-commerce email, was invisible to three of the four engines for this phrasing. This is the same split we documented between Google’s own two surfaces in AI Overviews versus AI Mode, now playing out across the whole field.
Finding 3: they do not cite the same sources
The brand list is only half the story. Each engine built its answer from different pages. Perplexity’s top cited source for the CRM answer was a small AI receptionist vendor’s blog, not any of the CRM brands and not a major publication. For the email question it leaned on niche roundups like emailvendorselection and mailtrap. Google’s AI Overview cited a blog called Saltbox. ChatGPT pulled from TechRadar and comparison sites.
That matters because the page that earns you a citation on one engine earns you nothing on another. Winning Perplexity can mean getting mentioned in a niche roundup that Perplexity happens to trust. Winning Google’s AI Overview means ranking in the sources it pulls from. There is no single page you optimize once. This is the citation-tracking job our reviews of Otterly.AI, AthenaHQ, and Peec AI all try to solve, and it is why they exist as separate per-engine tools.
What this means for your brand
- One engine is not the market. ChatGPT still sends the most referral traffic, but it now holds under half of AI assistant usage. If you only track ChatGPT, you are blind to the Gemini and Google AI Overview answers that a growing share of your buyers see. Track visibility on each engine, not an average. Our best GEO tools guide covers the ones that do this.
- The safe core is category giants. If you are HubSpot, you appear everywhere. If you are a challenger, your presence is decided per engine, so that is where the work is. A free AI brand visibility audit will show you which engines already name you and which do not.
- Optimize the sources, not just the site. Because each engine cites different pages, getting recommended means earning your way into the roundups, reviews, and comparison pages that each specific engine pulls from. That is closer to digital PR than to classic on-page SEO, and it changes how you split a GEO versus SEO budget.
The bottom line
The week’s headlines all point the same way. ChatGPT under 50 percent, new default models across the field, referral traffic splitting between engines. Our test is the ground-level version of that shift. Ask four AI engines the same buying question and you get four different answers, built from four different sets of sources. The brands that win are either too big to leave out or actively earning mentions on each engine separately. That means treating AI visibility the way you already treat search: measured continuously, per engine, and earned page by page. If you want a map of who names you today, start with our AI visibility tools roundup and our hands-on Profound alternatives test.