Why we separate “recommended” from “suggested”

The problem

Many simple AI visibility tools count a firm as visible if its name appears anywhere in a response. That collapses very different outcomes into one number. A firm explicitly recommended for the user's situation is in a different position from a firm included in a neutral list, a firm named in passing, or a firm mentioned in a cautionary context.

If those outcomes are counted the same way, a mention rate is reported as if it were a recommendation rate — and visibility is overstated.

Why it matters

A law firm can appear in an AI answer without being recommended.

For a law firm, the commercially meaningful question is rarely “were we named?”. It is “were we put forward to the person asking?”. A prospective client reading an answer responds differently to “I would recommend starting with Firm A” than to “firms operating in this area include Firm A, Firm B and Firm C”.

Example

Prompt

Which personal injury firms in Sydney would you recommend?

Response A

Several firms operating in Sydney include Firm A, Firm B and Firm C.

Response B

I would recommend starting with Firm A and Firm B.

Both responses answer the same recommendation-seeking prompt. In Response A the firms are listed; the system does not express a preference. In Response B the system explicitly directs the user towards two firms. FirmRanker would classify the firms in Response A as suggested, and Firm A and Firm B in Response B as recommended.

How FirmRanker handles it

Classification is applied per firm, per observation, and can be reviewed by a human against the preserved raw answer.

• Mentioned — Appears substantively but is not placed into an actionable consideration set.
• Suggested — Presented as a firm the user may consider.
• Recommended — Explicitly recommended or endorsed by the system.
• Source-only — Appears in source material but is not substantively surfaced.
• Negative / cautionary — Appears in a materially adverse or cautionary context.

What we still don't know

The boundary between a strong suggestion and a recommendation is not always clear-cut, and phrasing differs between AI systems. We are measuring how often automated classification agrees with human reviewers before relying on these categories for published findings. We do not yet know how much readers' choices differ between suggested and recommended firms.