Reading AI Reports: Mention, Position, Citation
What mention, position, recommendation and citation tell you in an AI visibility report. A reading order and common misreadings for B2B SaaS teams.
Short answer: An AI visibility report measures four separate things. Whether the brand appears in the answer is mention. Where it appears in the answer is position. Whether the engine recommends you is recommendation. Whether the engine links to your site is citation. Each one answers a different question and supports a different decision. When you read a report, check the engine set, the answer count and the margin of error first. Then read each signal per engine and against the previous measurement. A single composite score does not tell you what changed.
A B2B SaaS marketing lead who opens a report for the first time usually looks at one number: did the score go up or down? When it drops, the team wants to change the content plan. When it rises, everyone relaxes. Both reactions are usually early. Under the score sit four signals that move independently, and you cannot pick the right next step without knowing which one moved.
This post is not about how to build a report. It is about how to read one once it lands on your desk. For the building side, see our report template post. For the general logic of monitoring, see our AI visibility monitoring guide.
1. Four signals, four separate questions
The Botfusions AI Visibility Score is built from four signals. The weights are written on our methodology page: Mention 40%, Position 30%, Recommendation 20%, Citation 10%.
Mention
Whether the brand appears in the generated answer at all. It carries the highest weight because you cannot be chosen if you are absent from the answer. In a report it usually shows up as a mention rate: in how many of the measured answers does your brand appear?
What it tells you is whether the engine treats you as a candidate for this set of questions. What it does not tell you is whether it recommends you.
Position
Where the brand sits within the answer. Attention drops sharply as readers move down. A mention in the opening lines is worth the most.
Being fifth in a list of five and being the first recommendation look the same in a mention count. They do not have the same effect on a buyer's shortlist. Position lets you see that difference.
Recommendation
Whether the brand is recommended, listed neutrally, or mentioned with a caveat. A neutral-heavy mix means the engine knows you but will not recommend you.
There is also the source-only mention. Your name appearing in a sentence like "agencies listed on this site" is not a recommendation. In our 29 October 2026 measurement, 3 of our brand's 39 mentions were of this kind. That is why we read every answer that names the brand.
Citation
Whether the engine includes a direct link to your brand's domain. When a link is there, it is the most direct proxy for traffic that can reach your site. Not every engine links out in every answer, though.
Mention and citation are not the same thing. In the same 29 October 2026 measurement, Botfusions was mentioned in 39 of 476 answers, while botfusions.com was cited as a source in 71 answers. So there were answers where the engine read our site and never named our brand. Being used as a source does not mean your name made it into the answer.
2. Which decision each signal supports
| Signal | Question it answers | Where to look first when it is low |
|---|---|---|
| Mention | Does the engine see me as a candidate for these questions? | Comparison sites, list posts and directories the engine cites |
| Position | Where am I on the shortlist? | Your pages that answer the buyer's question in the first paragraph |
| Recommendation | Does it recommend me or just list me? | Caveats attached to you in answers, outdated pricing or integration details |
| Citation | Does the engine read my site and link to it? | Crawler access, indexing, sections that can be quoted on their own |
The third column is not a list of fixes. It is where to look first. The answer texts and the source list show why a signal dropped.
3. Common misreadings
Mistaking share of voice for mention rate
Share of voice is your share of all brand mentions across the prompt set. Mention rate is the share of answers you appear in. The two can move in different directions. If competitors drop out of answers, your share of voice rises while the number of answers naming you stays the same. When someone says "our share of voice went up", check the mention rate too.
Treating one run as a trend
Answers are generated fresh every time. OpenAI's API documentation states that outputs are non-deterministic by default. A single scan is a snapshot. Direction comes from a series of measurements taken with the same prompt set and the same engine set. We covered the causes of this variance in our post on why AI answers show your brand only sometimes.
Counting a difference inside the margin of error as change
According to our methodology, a scan of 200–300 answers carries a noise margin of roughly ±4–5 points per engine. If the difference between two measurements is smaller than that band, we do not count it as change. If an engine fails to answer part of the questions, the scan counts as a partial measurement and its numbers are not published. Look for that note in the report.
Comparing different engine sets
A score is comparable only against another measurement taken with the same engine set and the same depth. If a new engine was added to the report, the change in score may come from the set itself, not from your brand.
Reading a missing overview as a drop
Google states that AI Overviews are shown only when its systems determine they add something to classic Search. When no overview appears for a search, your brand's absence does not mean the engine dropped you. In our 29 October 2026 measurement, 2 Google AI Overviews searches returned no overview at all. We took those out of the denominator.
Missing spelling variants
A brand name can be written in different ways in answers. In our October 2026 agency research, spelling variants had undercounted one agency's mention share by 15 points. We fixed the counting rule and recalculated the numbers. When you read a report, ask which spellings of your brand were counted.
4. Reading order: a checklist
Read the report in the same order every time:
- Engine set and date. Which engines ran, and on which day? Is there a partial-measurement note?
- Answer count and margin of error. Are they written separately for each engine?
- Mention rate. Look per engine, not at the composite score.
- Difference from the previous measurement. Are the prompt set and engine set the same? Does the difference exceed the margin of error?
- Position. In the answers that name you, are you at the top of the list or at the bottom?
- Recommendation. Read the answers that name your brand. Recommendation, neutral listing, caveat, or source only?
- Citation and the source list. Which domains were cited? Which of your pages are there?
- Raw data. Are the prompts and answer texts attached to the report? We listed the fields that should be there in our report template post.
- Interpretation and next step. Are they written separately from the measurement?
If you are just starting to measure, read this list together with our first 30 days post before your first report.
5. When the composite score is useful
A composite score is good for telling leadership where things stand in one line. It is not enough to make a decision. The same score can hide a month in which mentions rose and position fell, or the reverse. Those two situations call for different work.
Our suggestion: read the score once at the top of the report, then go down to the four signals. Always choose the next step based on the signal that moved.
Limits of this post
The weights and definitions belong to our methodology. Other tools may measure different things under the same names. When you read another tool's report, ask for its definitions first. The examples in this post come from one measurement taken on a single day with agency questions. That measurement ran on ChatGPT, Gemini, Google AI Overviews and Google AI Mode. It did not measure position or positive and negative context. The picture in your SaaS category may be different.
