How CiteHawk works.

CiteHawk asks the AI platforms the questions your buyers ask, records what comes back, and turns the record into numbers you can track. This page explains what is collected, how often, and how each score is built from it.

What gets measured

The unit of measurement is a prompt: one buyer question, asked of one AI platform, on one collection run. A prompt looks like something a real customer would type, not a keyword. CiteHawk sends each of your tracked prompts to every AI platform your plan collects, then stores the full answer that comes back along with any sources the platform cited.

Each stored answer is read for three things:

Whether your brand is named, and whether your competitors are named alongside you.
Which sources the platform cited, so you can see the domains that shaped the answer.
How your brand is described, which is what the Brand Health score reads.

Nothing is inferred from a search ranking or a proxy signal. Every number in CiteHawk traces back to answers that were actually collected. See the receipts principle below.

Collection cadence

Collections run weekly, every Monday. Each run asks every tracked prompt of every AI platform your plan collects, so week over week you are comparing like with like.

Weekly is deliberate. AI platforms vary their answers from one ask to the next, so a single collection proves very little. A steady cadence over the same prompt set is what turns a noisy signal into a trend you can act on.

On top of the weekly run, you can start the same collection on demand, for when you have shipped a change and do not want to wait until Monday to see whether it moved anything. Each month your plan carries a run budget worth four full sweeps; the weekly sweeps draw from it and always run, and on-demand runs use whatever remains. See Billing and plans for how the budget works.

AI platforms per plan

Starter and Pro collect 7 AI platforms: ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Copilot, and Google AI Overviews. Growth and Agency add Google AI Mode, for 8. Grok is a paid add-on on every plan.

A free trial runs 8 platforms, the widest included roster, at Starter prompt volume, so the first report shows it before you choose a plan. A trial does not collect Grok.

The platforms split into two kinds. Conversational platforms answer in prose and are asked directly. Search platforms are the AI answers that appear above ordinary search results. Both are collected the same way and both count toward your scores, but they behave differently, and Rankings breaks your numbers out per platform so you can see where they disagree.

The scores

CiteHawk leads with two 0 to 100 scores. Visibility asks whether AI recommends you to buyers. Brand Health asks whether AI describes you accurately and positively. Underneath them sit the component rates the scores are built from.

Visibility score

Your Visibility score (workspace) is a single 0 to 100 number built from your own tracked prompts. It is not the same thing as the AI Index score (public leaderboards), which ranks a whole category from a separate prompt set under a separate formula, published in the Index methodology. Two questions, two formulas, and both of them published in full.

Visibility reads discovery and competitor prompts only, the questions that do not name you. It has two parts: Presence, worth up to 70 points, and Authority, worth up to 30.

Visibility (0 to 100) = Presence + Authority

Presence  = 70 x (discovery mention rate) ^ 0.6
Authority = 30 x min(authority citation rate / 25%, 1) ^ 0.6

The 0.6 power is a response curve, not a bonus. It lets real presence read honestly without moving the ceiling: a mention rate of 100 percent still scores the full 70 points, while a mention rate of 25 percent scores 30.5 rather than 17.5. Authority works the same way against a ceiling of 25 percent, which is a high authority citation rate in practice, so anything at or above it earns the full 30 points.

Both rates share one denominator: the analysed answers to your discovery prompts. A prompt you have excluded leaves the count entirely. An answer with no analysis row, because analysis failed or is still pending, is out of the numerator and the denominator together, so a rate can never exceed 100 percent.

The discovery mention rate is the share of those answers that name your brand. The authority citation rate is the share that cite your own domain, or an authority or news page that references your brand.

Movement is reported as a delta against your previous score rather than folded into the score itself, so a Visibility score of 60 means the same thing whether you are rising or falling. The score is the headline; the receipts are the product, and every movement traces back to the specific answers that changed.

Mention rate

The fraction of AI answers that name your brand at least once, out of all answers collected for your tracked prompts. A mention rate of 0.6 means you appeared in 60 percent of answers.

Mention rate is the presence number: the simplest measure of whether AI knows you exist for a given question. It ignores position and volume, so one mention in an answer counts the same as five.

The percentage of brand mentions across a set of AI answers that belong to you. If AI platforms name brands 200 times across your tracked prompts and 30 of those mentions are you, your share of voice is 15 percent.

Share of voice is a market-shape number. Mention rate answers whether you are present; share of voice answers how much of the room you take up. Two brands can both appear in 8 of 10 answers while one leads every list and the other trails at the bottom. Read the two together.

Citation rate

The fraction of AI answers that cite your own website as a source. It measures whether AI platforms are reading and crediting your site, not merely naming you.

An AI platform can name you from third-party sources alone: reviews, directories, comparison articles. Citation rate isolates the cases where your own pages made the reading list. A healthy mention rate with a near-zero citation rate means AI talks about you using other people’s words.

The citation rate shown on your dashboard counts your own domain only. The Authority component of the Visibility score uses a wider count: it also credits an answer that cites an authority or news page referencing your brand, which is why the two numbers differ.

Brand Health score

Brand Health is the second 0 to 100 score, and it reads branded prompts only, the questions that name you. It is Sentiment, worth up to 50 points, plus Accuracy, worth up to 50.

Brand Health (0 to 100) = Sentiment + Accuracy

Sentiment = 50 x (average sentiment + 1) / 2
Accuracy  = 50 x (confirmed-accurate share of your brand facts)

Average sentiment runs from -1 to 1 and maps onto 0 to 50 linearly, so a neutral read scores 25. Accuracy maps the share of your brand facts that AI platforms state correctly onto the other 50. Until your brand has verified facts to check against there is no accuracy rating at all, and Brand Health is the sentiment component scaled to 100 on its own.

The receipts principle

Every number in CiteHawk opens to the real AI answers behind it. Click a score, a rank, a mention, or a source, and you reach the collected answers that produced it, with the platform that gave them and the date they were collected.

This is a design rule rather than a feature, and it has consequences:

Answers are archived, not summarized away. The text the platform returned is kept, so a number can always be re-derived from its evidence.
The platform and the date are stamped on every answer, because an answer without them is an anecdote.
Nothing is modeled or estimated into existence. If a prompt was not collected on a platform, CiteHawk shows a gap rather than filling it in.

The practical payoff is arguing from evidence. When a number moves, you can show the answers that moved it, which is what makes AI visibility reportable to someone who was not in the room.

Next: read what each surface does, or see how plans are metered.