Ask four AI assistants which business software to buy and you will mostly get four different answers. That is the headline finding of a new study by AI visibility platform Envoyra, which asked ChatGPT, Claude, Gemini and Perplexity the same buyer-style questions every week for four weeks and found that the assistants named the same top brand in only one of five software categories. Search Engine Land reported the study on October 22.
Envoyra collected 2,720 answers in total, and the disagreement was not the only instability it found. In 79% of the question-and-assistant combinations it tracked, the list of recommended brands changed at least once across the four weekly runs, even though the questions stayed the same.
Envoyra’s Method: 170 Questions, Four Assistants, Four Weeks
Envoyra ran 170 buyer-style questions per assistant per week, in four Sunday runs from September 13 to October 4, 2026. The questions covered five categories: meeting assistants, customer support AI, CRM for small business, marketing automation and AI visibility tools. The models were ChatGPT (GPT-4o, with web search), Claude (Sonnet 4.6, no web search), Gemini (2.5 Flash, no web search) and Perplexity (sonar-pro, with web search) — a mix of API versions and settings that Envoyra itself notes may differ from what people see in the consumer apps.
A separate test, run from October 7 to 9, asked the assistants about 30 of the brands by name, producing 720 recognition answers. Envoyra does not publish its exact questions, and it competes in the AI visibility category that the study covers — two limits it discloses in the report, and worth keeping in mind when reading the numbers.
The Assistants Disagree About Who Leads Each Category
Zendesk, in customer support AI, was the only brand to rank first on all four assistants. Everywhere else, the leader depended on which assistant a buyer asked. HubSpot led CRM on ChatGPT, Claude and Perplexity, while Gemini favoured Zoho CRM. ChatGPT picked Otter.ai for meeting assistants, Claude and Perplexity picked Fireflies.ai, and Gemini picked Zoom. In marketing automation, ChatGPT favoured HubSpot, Claude favoured Klaviyo, and Gemini and Perplexity favoured ActiveCampaign.
The sharpest split sat inside the AI visibility category itself. ChatGPT and Gemini favoured Semrush, while Claude and Perplexity favoured Profound. Claude recommended Profound in 89 of 96 answers; Gemini never mentioned it at all — a zero Envoyra double-checked by re-running the questions on October 9. Claude also named Otterly.AI and Peec AI 75 times each, while Gemini named neither in any answer.
Below The Leaders, The Lists Keep Changing Week To Week
The top of each category was steadier than the rest of the list. HubSpot appeared in 305 of 368 eligible CRM answers, with weekly counts of 77, 76, 75 and 77, and Intercom was named in 62 of 96 customer support answers across the period. Underneath those leaders, though, the shelves kept moving: of the 680 question-and-assistant pairs Envoyra tracked, only 144 returned the same brand list in all four weeks. The other 536 — 79% — changed at least once, with a brand appearing one week and gone the next, or a new name joining.
Envoyra checked that pattern against earlier research and found it held up. SparkToro and Gumshoe’s study, which covered ChatGPT, Claude and Google’s AI, likewise found the same brand list coming back fewer than one time in 100, while a handful of leading brands kept turning up. That earlier work is part of a growing body of evidence — including the citation research behind the paper that found OpenAI’s API matched only 26% of the pages ChatGPT cited — that a single AI answer is a sample, not a ranking.
Being Known Is Not The Same As Being Recommended
The recognition test produced the study’s strangest results. Every assistant correctly described Keap in all 24 of its recognition answers, yet Keap was recommended in just 3 of 400 CRM answers. Profound showed the mirror image: recommended in 136 of 384 category answers, but correctly identified by name in only 5 of 24 tests. Fathom (recommended in 134 of 368, recognised in 8 of 24) and Otterly.AI (119 of 368, recognised in 13 of 24) followed the same pattern.
Many of the misses were name collisions rather than inventions. Asked about Fathom, some assistants described a financial-reporting company with the same name; asked about Profound, some described Profound UI, a developer tool from Profound Logic; Otterly.AI appears to have been confused with Otter.ai. For brands tracking their presence in AI answers — the work covered in our guide to how brands track visibility inside ChatGPT and Gemini — recognition and recommendation are plainly two different numbers, and a brand can score high on one and low on the other.
Why It Matters
AI visibility is becoming a budget line, and this study is a warning about how it gets measured. A brand that checks one assistant, once, learns almost nothing: the same question can produce a different shortlist next week, and a rival that leads on Claude may be invisible on Gemini — as Profound’s 89-to-zero split shows. That instability sits alongside the scale findings in Semrush’s study of AI Overviews on commercial keywords: AI answers are now in front of most buyers, but no single answer is a stable place to stand. For software companies, the practical reading is Envoyra’s own — track how often you appear across several assistants over several weeks, fix the directory and profile pages the assistants read, and treat any one-off ranking claim from a tracking tool with suspicion, including the limits of this study itself: unpublished questions, API models rather than consumer apps, and a researcher that sells visibility services in the very category it measured.
Background
Envoyra is a Washington, DC-based AI visibility platform founded by Golara Serio, and its Fall 2026 report is the first edition of what it frames as the “Invisible Shelf” — the short set of brands AI systems actively recommend to buyers. The study lands in a field that has been building through 2026, from SparkToro and Gumshoe’s January finding that AI recommendation lists almost never repeat, to a steady run of visibility research from Semrush, Profound and other tracking vendors whose own products are now being measured by the assistants they monitor.
Source: Envoyra AI Recommendation Report, Fall 2026; Search Engine Land
0 Comments