Measurement is only useful if it ends in a to-do list. Recommendations is that list: what to do next for this brand, most important first, written from its own answers rather than from a generic GEO checklist.
What a recommendation looks like
Each recommendation has:
- A type, so you know whose desk it lands on.
- A one-line reason: what in your results prompted it.
- Numbered steps: what to actually do.
| Type | What it covers |
|---|---|
| Content | Pages to write or improve so your site answers the questions you're missing from |
| Sources | Third-party sites to get onto or corrected: the reviews, roundups and listings AI cites |
| Positioning | How you're described: themes to reinforce or counter |
| Technical | Things that help AI read your site: structure, markup, crawler access |
They're drawn from everything else brandflare measures: the questions where you're not named, the sources the engines cite, the false claims they repeat, the themes in how they describe you, and where competitors are named instead.
Working the list
Recommendations has three tabs:
- To do: what's open, most important first.
- Done: what you've completed.
- Dismissed: what doesn't apply.
Mark a recommendation done when the work ships, and dismiss the ones that don't fit your brand. Dismissing isn't failure: brandflare doesn't know you've decided not to compete on price, and clearing those out keeps To do honest. Editors, admins and owners can change a recommendation's status; viewers can read the list.
The sidebar shows how many are open beside the section's name, and a new brand's recommendations are ready as soon as its first check finishes.
What to do next, on the overview
Your brand's overview has a shorter, sharper version: What to do next ranks false claims, missed questions, negative stories and recommendations together, with a button for each. Think of it as today's list, and Recommendations as the backlog.
Picking what to do first
The list is ordered by importance, but you know your own constraints. A useful way to sort it:
- False claims first. A wrong price or a "discontinued" label costs sales every day it stands. These aren't recommendations at all; they're in Answer Accuracy.
- Then the near misses. Questions where you're named by some engines and not others, and sites that are cited often and rarely mention you. You're already close, so the work pays back fastest.
- Then the blanks. Questions where no engine names you need new material and new coverage, which takes longer.
- Technical items in parallel. They're usually a one-off job for a web team and don't compete with content for anyone's time.
Knowing whether it worked
Ship the work, mark it done, and keep checking. The weekly checks tell you what happened next: a cell on Questions & Prompts that turns from Not named to 3rd, a theme that fades, a source that starts naming you. Engines that search the web respond within weeks; changes to what a model has memorised arrive with its next release. The realistic timeline is in how long does GEO take to work?
For the playbook behind the recommendations, see how to improve AI visibility.