Ask an AI assistant to recommend a product twice and you may get two different lists. Ask another assistant and the category leader may change too. That does not automatically mean your brand gained or lost ground.

The list can change while the leaders stay put

Envoyra’s Fall 2026 research tested 170 software-buying questions across four assistants for four weeks, collecting 2,720 answers. Of 680 question-and-assistant pairs, 536 returned a different brand list at least once. That is the 79% figure: not 79% of brands disappearing, or 79% of answers being wrong.

At the same time, some leading brands stayed remarkably steady. HubSpot appeared in 77, 76, 75, and 77 eligible CRM answers across the four weekly runs. A changing shortlist and a stable aggregate leader can both be true.

You are comparing systems, not just brand names

The report used web search for ChatGPT and Perplexity, but not for Claude or Gemini. That matters. A search-enabled answer can draw on recently retrieved pages; an answer without search has a different information path. These were particular API models and settings, not a universal ranking of the consumer apps.

There is also variation within a system. Generating an answer involves choices about what to include, which alternatives fit the question, and how much room to give each one. A borderline recommendation can move in or out even when the strongest candidates keep appearing.

My interpretation is that teams often put three different effects in the same bucket: ordinary response variability, differences between platforms, and genuine change over time. Each needs a different comparison.

A single missing mention is a question, not a conclusion

Suppose your brand appears in one answer this week and not in the next. That is worth recording. It is not enough to conclude that a competitor’s new page displaced you.

A more useful view keeps the questions and settings consistent, samples repeatedly, and separates results by platform. Then you can ask whether your share of answers has actually changed, whether the same competitor is repeatedly preferred, and which reasons the answers give.

The reasons are where this gets interesting. A brand may still appear just as often but lose on reliability, ease of use, or price. That is a different problem from disappearing altogether, and a mention count alone will miss it.