Quickstart
From a bare domain to a first reading you can trust.
Setup takes a few minutes. Getting a reading you can act on takes as long as the models take to chat — which is why the last step here is about waiting, and what to look at while you do.
Add the brand
Give Recomma your domain and the market you sell into. It reads the site and works out what you sell, who you sell against, and the language your buyers use for it.
Two things come out of that read, and both are proposals rather than decisions:
- Prompts — the questions Recomma thinks your buyers ask.
- Competitors — the brands it expects to be named alongside you.
Nothing is tracked until you approve it. A proposed prompt costs nothing while it sits there; it starts costing when you accept it, because accepting it means running it on every model, every cycle, from then on.
Fix the prompt set before anything runs
This is the step that decides whether the numbers are worth reading, and it is the one people skip.
A prompt is a question a buyer would actually type. Not a keyword — models answer questions, so a bare noun phrase measures nothing:
| Instead of | Write |
|---|---|
project management software | What's the best project management tool for a 10-person agency? |
ahrefs pricing | How much does Ahrefs cost compared to Semrush? |
seo tools | Which SEO tool should I use if I mostly care about backlinks? |
Three things worth doing before you accept the set:
- Cover the whole decision, not just the top of it. Recomma marks each prompt with a funnel stage — awareness asks what exists, consideration compares options, decision is ready to choose. A set that is all awareness tells you about category presence and nothing about whether you win the comparison.
- Delete the ones you do not care about. Every prompt you keep is sampled on every model on every cycle. A set of forty questions you will act on beats two hundred you will not read.
- Group them into topics. A topic is the unit you will later filter the whole Overview by, so it is worth naming them the way your team already talks about the product.
Start smaller than you think
Coverage is easy to add later and expensive to unwind: the history you build is per prompt, so deleting one in a month throws away a month of its readings.
Check the competitors
Recomma proposes the brands it saw the models naming next to you, and you can add any it missed. This matters more than it looks: share of voice and position are both measured against this list. A rival that is not on it does not dilute your share, so the figure reads higher than the market really is.
Let it sample
Each cycle, every tracked prompt is asked on every model you have enabled, and each chat is stored whole. A project runs daily or weekly — daily if you are working actively on visibility, weekly if you are watching a trend.
The Overview will start showing figures after the first pass, but the first pass is not yet a reading. A percentage measured over a handful of chats moves several points on a single one, which is why Recomma shows you the sample size and a confidence interval beside every headline figure, and greys the figure out below twenty chats.
Read the Overview
Once the sample is real, the Overview answers four questions at a glance:
- Are we named? — Visibility, and its trend against the category leader.
- How much of the conversation is ours? — Share of voice, in the Rankings table.
- How is it spoken about? — Sentiment.
- What is the chat built on? — Top domains, and what kind of evidence they are.
- How much of the market is that? — the Market panel, once a keyword pool has been built for the project.
Then go and change something
Visibility is a symptom. The cause is which pages the models trust on your subject and whether those pages mention you.
Sources ranks the sites the chats actually leaned on. Opportunities narrows that to the ones being trusted on your subject while saying nothing about you — the cheapest ground to go and be present on. Impact reads the figure back afterwards, so a change you made has a before and an after rather than a hunch.
Two of those pages need no chat at all. Site reads your own pages and says what the markup is missing, and Keywords prices the market your questions are drawn from and says which part of it no question measures yet — that one waits for the first pool build, which runs on the schedule.