How a HiBot audit measures AI visibility
This page is the reference version of our method. It covers what we ask, which engines we ask, how each answer is scored, and how the scores roll up. You are welcome to cite it.
1. How answers are captured
Each audit starts from a set of questions written the way customers ask them, which the brand reviews and approves. Trained specialists then ask every question by hand on every engine in the plan. Each session is fresh and logged out, and set to the brand's target country and language. The specialist saves the answer word for word, along with the sources the engine cites. Those capture conditions stay the same on every repeat audit, so trends compare like with like. On a repeat audit the approved question set is locked and reused.
Scoring reads the saved text only. We never send the question to an engine again to score it, because a second query can return a different answer from the one the customer would have seen.
2. Which AI engines each plan covers
| Plan | Engines | Runs per question |
|---|---|---|
| HiBot Pulse | ChatGPT, Perplexity, Google | 1 |
| HiBot Panorama | ChatGPT, Perplexity, Google, Claude, Grok | 2 |
"Google" means the AI-generated answer shown in Google Search. No plan currently covers Gemini (the standalone app), Microsoft Copilot, Meta AI, DeepSeek. We left Copilot out on purpose: it runs on OpenAI models, so its answers overlap heavily with ChatGPT, and Grok adds a more distinct point of view to Panorama.
3. The six ANSWER categories
ANSWER follows the customer decision journey, from first consideration through purchase and loyalty. Weights set how much each category counts toward the headline score. Question counts set how reliably each one is measured, so the two are deliberately separate.
| Category | Journey stage | What it measures | Weight | Pulse questions | Panorama questions |
|---|---|---|---|---|---|
| Appearance | Consideration | Whether AI names you at all when a customer asks an open question about your category. | 25% | 7 | 12 |
| Nomination | Evaluation | Whether AI puts you forward for a specific use case or customer situation. | 15% | 4 | 8 |
| Showdown | Evaluation | How you fare when a customer asks AI to compare you with a named competitor. | 20% | 5 | 10 |
| Worth | Purchase | Whether AI states your price and value correctly at the moment of decision. | 10% | 3 | 5 |
| Expertise | Evaluation | Whether AI describes your features and capabilities correctly. | 15% | 4 | 9 |
| Reputation | Experience and loyalty | Whether AI reflects your standing, reviews, and trust signals correctly. | 15% | 2 | 6 |
4. How each answer is scored
Every saved answer gets a score from 0 to 4. Open questions, the ones that do not name the brand, use this scale, where being mentioned is the first thing that counts:
| Score | Label | Meaning |
|---|---|---|
| 4 | Recommended first | The brand is the primary recommendation, or wins the head-to-head. |
| 3 | Recommended | The brand is recommended or presented favorably, but is not the clear leader. |
| 2 | Mentioned | The brand is listed or referenced without an endorsement. |
| 1 | Absent, rivals named | The brand is left out and competitors or alternatives are named. |
| 0 | Absent | The brand is not mentioned at all. |
Showdown, Expertise, and Reputation questions name the brand on purpose, and the answer names it back almost every time. On those questions a mention earns nothing, and the score depends on what the answer says:
| Score | Label | Meaning |
|---|---|---|
| 4 | Right and favorable | Accurate, specific, and favorable, or the brand wins the head-to-head. |
| 3 | Right, positive | Accurate and positive, but thinner or not the clear winner. |
| 2 | Right, generic | Accurate but generic or heavily hedged, or an even split. |
| 1 | Unknown or beaten | The engine says it has no reliable information about the brand, or recommends a competitor instead. |
| 0 | Wrong company | The engine mistakes the brand for another company or gets it fundamentally wrong. |
Audits scored before October 16, 2026 used the first scale for every question, and their reports keep the scores they were delivered with.
If an answer names the brand but gets a material fact wrong, or frames the brand negatively, the score is capped at 2 however prominent the mention is. The anchors also adjust to the category. In Showdown, a 4 means the brand won the comparison and a 1 means the engine chose the competitor.
5. How scores roll up
- Question score: the average of the engine scores for that question. On Panorama, each engine's two runs are averaged first, which rewards answers that stay consistent.
- Category score: the average question score in the category, multiplied by 25, which gives a number from 0 to 100.
- HiBot Score: the weighted blend of the six category scores, using the weights in section 3.
Scores fall into five bands: Dominant (80 and above), Strong (60 to 79), Emerging (40 to 59), Marginal (20 to 39), and Invisible (below 20).
6. Share of voice, citation rate, and accuracy
These three metrics sit beside the HiBot Score. None of them changes it.
Share of voice is the number of saved answers that name the brand, divided by the number of answers that name the brand or at least one tracked competitor, counted on the questions that name no company. An answer that names both counts on both sides. Registered aliases count as mentions, so "Delta" counts for Delta Air Lines.
Citation rate is the share of scored answers that cite a page on the brand's own website.
Accuracy rate is the share of answers that mention the brand without a factual error made by the model. Every claim an engine states as fact goes into a claims log and is checked against the brand's own site. Errors repeated from a third-party source, or from the brand's own pages, are labeled as such and do not count against accuracy.
Common questions about the method
Which AI platforms does a HiBot audit cover?
HiBot Pulse covers ChatGPT, Perplexity, Google. HiBot Panorama covers ChatGPT, Perplexity, Google, Claude, Grok. No HiBot plan currently covers Gemini (the standalone app), Microsoft Copilot, Meta AI, DeepSeek.
How does HiBot measure share of voice in AI answers?
Share of voice is the number of saved answers that name your brand, divided by the number of answers that name your brand or at least one of the competitors you track. It counts only questions that name no company, because a company named in the question almost always appears in the answer. An answer that names you and a competitor counts on both sides. Any brand alias you register counts as a mention.
What is the ANSWER framework?
ANSWER is HiBot's six-category method for scoring AI visibility: Appearance, Nomination, Showdown, Worth, Expertise, and Reputation. Each category maps to a stage of the customer decision journey and is scored 0 to 100. A weighted blend of the six gives the 0 to 100 HiBot Score.
Are HiBot audits automated?
Answer capture is not. Trained specialists ask every question by hand in fresh, logged-out sessions set to the brand's target country and language, and save the answer word for word. AI then scores those saved answers only. HiBot never re-queries an engine to score it.
Longer reads: How an ANSWER Audit Is Run, How an ANSWER Audit Is Scored, and the ANSWER whitepaper.