Some of the questions in a HiBot audit name your company on purpose, like "How does Acme compare to Rival?" or "Is Acme a trustworthy vendor?" Those questions test whether AI gets you right once a customer already knows your name. They were never meant to test whether AI knows you exist.
The trouble is that when a question names you, the answer almost always names you back. Our original rubric gave credit for that. Someone reading one of our customer reports pointed this out. We checked the data, and they were right. So we have changed two things.
What Was Wrong
Every ANSWER audit asks two kinds of questions. Appearance, Nomination, and Worth questions are open: "What is the best recruiting software for a small agency?" Your name is not in the question, so the answer only mentions you if the engine thinks of you on its own. Showdown, Expertise, and Reputation questions name you, because they test head-to-head standing and accuracy.
We were counting mentions across both kinds. On the named questions, a mention is close to guaranteed, so the totals looked better than reality. Our own audit of HiBot shows how much. Across all 25 questions, the engines mentioned HiBot in 47% of answers. On the open questions alone, it was 5%.
Scoring had a smaller version of the same problem. The rubric gave 2 out of 4 for a neutral mention. On a named question, an answer could say "I could not find reliable information about this company" and still earn a 2, sometimes more, just for discussing the brand. An answer that confused the brand with a different company with a similar name also scored 2.
Change 1: Mention Metrics Count Open Questions Only
Brand mention rate, coverage by engine, and share of voice now count only questions that do not name you or any competitor you track. The same rule covers competitors. A Showdown question that names a rival guarantees that rival a mention too, so those questions are left out of share of voice entirely.
These numbers now answer the question they were always meant to answer: when a customer asks AI for options without naming anyone, how often do you come up? On our own audit, share of voice fell from 56% to 14%. That is a less flattering number, and it is the more accurate one.
Change 2: Named Questions Are Scored on What the Answer Says
When the question names you, being mentioned earns nothing on its own. The score depends on whether the answer gets you right and how it treats you.
| What the answer says | Original rubric | New rubric |
|---|---|---|
| Accurate, specific, and favorable, or you win the comparison | 4 | 4 |
| Accurate and positive, but thinner or not the clear winner | 3 | 3 |
| Accurate but generic or heavily hedged | 2 | 2 |
| "I could not find reliable information about this company" | 2 to 4 | 1 |
| A competitor is recommended over you | 1 | 1 |
| The engine mistakes you for a different company | 2 | 0 |
The open questions keep the original rubric, where a mention still matters, because there the engine chose to name you.
Why This Is Better
It measures what your customers run into. Someone who has never heard of you will not type your name into ChatGPT, so presence on open questions is the visibility that brings in new customers. And when someone does ask about you by name, an engine that says it has never heard of you, or describes a different company, is a problem. The new rubric scores it as one.
It is also fair across companies. Every brand gets the same mix of open and named questions, so no one gains from being named more often than anyone else.
Headline scores move less than you might expect. On our own audit the HiBot Score moved by about a point, because the categories that name you were never the whole score. Brands that AI describes well will see little change. Brands that AI does not know yet, or keeps confusing with someone else, will see lower Showdown, Expertise, and Reputation scores. Those scores now reflect what the engines say.
What Has Not Changed
The 6 ANSWER categories and their weights are the same. Your question set stays locked, so trends still compare like with like. We still capture answers in fresh, logged-out sessions set to your market, and we still score only the stored answers. We never re-ask the engines.
Your Existing Reports
Audits delivered before October 16, 2026 keep the score they were delivered with, calculated under the original rubric. We have not changed any past report.
If you want an earlier audit scored under the new rubric, email us and we will rescore it. It takes no new capture: we apply the new rubric to the answers already stored in your audit, so you get a clean comparison of the same answers under both methods.
