How to compare what AI assistants recommend: you against your competitors

The dash that was lying

For two weeks our own reports showed a small grey dash in most boxes of the grid. A dash was meant to mean the assistant named nobody. It did not mean that. It meant the assistant named nobody on your list. The names were extracted by a piece of code we had built and then, embarrassingly, never switched on.

When we switched it on, the dashes filled with names. In our own category the company named most often was one I had never come across. It was not on our list because we did not know it existed, which is the whole point of looking.

So list your competitors, then count everyone else too. If you only round up the usual suspects, you only find the usual suspects.

What a fair comparison needs

The same questions for you and for them. Questions that name nobody, because an assistant asked about a named company will talk about that company. Several assistants, because they disagree more than you would expect. And the answers kept, word for word, so when a competitor wins a question you can read the sentence it won with and the pages it was cited from.

What the assistants recommend for this, as of 22 September 2026

We asked ChatGPT, Gemini, Perplexity and Claude these questions on 22 September 2026. Their descriptions, summarised, not our review:

• Spyglasses Product Tracking: monitors visibility for specific products, measures recommendation share against named competitors and shows the sources the AI trusts.
• Friction AI: tracks the prompts behind product recommendations across recognition, consideration and purchase intent.
• Allmond: a competitive intelligence toolkit that tracks "Share of Model" against competitors.
• 42A: competitive intelligence across large language models, tracking mentions and position changes.
• RecoScope: benchmarks product recommendations across AI models from commercial prompts.
• Visor: audits what language models know about your product. One answer put it at $200 per audit.
• Seeto Feature Comparison: builds side-by-side feature matrices from competitors' sites, with a source for every claim.
• Answer Audit, which is us: 25 buyer questions written without naming you, five assistants, every answer kept, and everyone named counted, listed or not. Free check; full report $29 once.

Alerts when a competitor is named and you are not

For alerts as they happen, the assistants named Mentionary, Trakkr and Presenc AI, which let you set an alert for exactly that: a competitor named where you are absent. We do it monthly instead. Monitoring re-asks the same questions each month and shows every answer you gained and every one you lost. Most weeks there is nothing to alert about, so I think monthly is the honest cadence, but I would be glad to be argued with.