Evolution: track how AI talks about your brand over time, scan by scan
A single measurement is a snapshot. Evolution joins scans into trend lines, for your brand and for each competitor, and helps you tell a real change from the normal variation in AI answers.
Short answer
To track AI visibility over time, you compare scans run with the same questions, the same engines and the same market. Evolution shows four trend lines (share of voice, AI visibility, sentiment and position) for your brand and for each competitor, one color per brand. Every rate comes with its sample, as "k of n", so a small change over a few answers is not read as news. Scans are started by a person, on a cadence you set with the team.
What Evolution shows
Each point on a line is one scan. For each scan, four metrics are plotted, for you and for every competitor you follow:
| Metric | What it measures | What to look for |
|---|---|---|
| Share of voice | Your share of AI mentions against your competitors. | Whether yours went up or a rival's went down: they are not the same. |
| AI visibility | How often the engines name you in their answers. | Whether the change repeats across several engines or sits in one. |
| Sentiment | How you are described, from 0 to 100: 70 or above is positive, 40 to 69 is mixed, below 40 is negative. | Whether you crossed a band or only moved within it. |
| Position | Where you appear when the answer is a list. | Whether you gain places against a specific rival. |
Seeing your competitors on the same chart changes the reading. A drop in your share of voice can come from a rival appearing more often, with nothing changed in how AI sees you. With one line and one color per brand, that is visible at a glance. And in the Command Center you can pick any past scan to see the full summary of that moment.
Why you compare with the same questions
A trend only compares something if the measurement stays the same. If you change the questions between one scan and the next, the line mixes two things: what changed in AI and what changed in your measurement. Adding ten purchase-intent questions where your brand is weak lowers your visibility even if no engine changed a word.
That is why it pays to keep the Prompts Library set stable: same questions, same engines, same country. If you add new questions, treat that scan as a new starting point.
There is a second rule: an engine that did not answer does not count as zero. If an engine fails during a scan, that gap is not recorded as "answered and didn't name you". Without that rule, a provider's technical outage would show up on your chart as a drop for your brand.
Real change or noise: how to read "k of n"
AI answers vary: the same question can return different lists on two tries. That is why every rate is shown with its sample. A 50% that comes from 5 of 10 answers is not the same as one that comes from 50 of 100.
An example to calibrate your eye: going from 3 of 10 to 5 of 10 is a change that can happen by chance; going from 30 of 100 to 50 of 100 is much harder to explain that way. Before you react to a movement, check four things:
- The sample size: the more answers, the more reliable the difference.
- Whether it repeats across engines: a change in a single engine can come from an update to that model.
- Whether it holds in the next scan: an isolated jump that returns to its level was variation.
- Whether it lines up with something concrete: a launch, a campaign, a media story or a competitor's move.
It also helps that Refractia reports chat engines and AI search separately. A chat engine reflects what the model knows; AI search reflects what the web ranks and the engine paraphrases. A change in one and not the other already tells you where to look for the cause.
How often to measure
Refractia does not monitor your brand continuously. Scans are started by a person, on your team or at Refractia, and each one is a deep measurement: identity, questions, five engines and every metric. It is a diagnostic instrument, not a dashboard that refreshes every day.
You set the cadence with the team, based on what you want to observe. It makes sense to measure around the moments that can move the answer:
- Before and after publishing content or running a press campaign.
- When you launch a product or enter a new market.
- When a competitor makes a big move.
- On a regular schedule that gives you a baseline.
What to do when a line moves
Evolution tells you that something changed. To find out what changed, cross it with the other modules:
Performance by Topic
Which topic in your category you gained or lost ground in.
Influence Sources
Whether sources AI cites about you appeared or disappeared.
AI Perception
Whether the adjectives and traits used to describe you changed.
Prompts Library
Each engine's answers, word for word, question by question.
Frequently asked questions
How do I track changes in what AI says about my brand?
By comparing scans run with the same questions, the same engines and the same country, and reading every change together with its sample. Evolution plots share of voice, visibility, sentiment and position for your brand and your competitors.
How often should I measure AI visibility?
It depends on what you want to observe. Measure before and after specific actions, such as content, press or launches, and keep a regular schedule as a baseline. The cadence is set with the team.
Does Refractia monitor my brand every day?
No. Scans are started by a person, and each one is a deep measurement across five engines. Refractia is a diagnostic instrument, not a daily tracker.
Why did my AI visibility drop if I didn't change anything?
A competitor may have gained presence, the question set may have changed, an engine may have updated its model, or the difference may be small relative to the sample. Reading the "k of n" and seeing your competitors on the same chart helps separate those causes.
Can I see how my competitors' AI visibility changes over time?
Yes. Every metric is plotted for your brand and for each competitor, with one color per brand.
Keep reading
Start by knowing where you stand today
Without a first measurement there is no trend to read. In the free diagnostic we show you what ChatGPT, Claude, Gemini, Google and Meta AI say about your brand and your competitors. No commitment: we reply within one business day.