Recommendations
A ranked queue of moves derived from your own data, each with impact, effort, rationale and action items.
Recommendations is the prioritized work queue. Each card is one move, argued from measured evidence and broken into steps you can assign.
Preview. The generator that builds recommendations from the gap detectors is not yet running, so this screen renders a sample queue. For work you can act on today, use PR Opportunities, which is live on your data. See Product status.
Anatomy of a recommendation
Title — the move in one line: "Add FAQ schema to the pricing page."
Rationale — the evidence, with numbers: "ChatGPT cites competitor pricing pages 3× more often; all of them ship FAQPage markup while /pricing has none."
Impact — High, Medium or Low. Expected effect on visibility, weighted by how much of your answer volume the affected prompts represent.
Effort — Quick win, Medium or Deep work. Realistic implementation cost.
Type — how to read the item:
| Type | Meaning |
|---|---|
| Alert | Something is actively going wrong. Look now. |
| Warning | A negative trend that will cost you if ignored. |
| Opportunity | An unclaimed gain. |
| Win | Something worked. Confirmation and a pattern worth repeating. |
Action items — the concrete steps, usually three, ending with re-running the audit so the effect is measurable.
Analysis — the longer explanation: what was observed, across which prompts and engines, and why this fix should move the number.
Prioritizing
Filter by impact and by type, then order by impact against effort. The queue rewards a simple discipline:
- High impact, Quick win — do these immediately. There are rarely many, and they are usually structural (schema, headings, a missing comparison section).
- High impact, Deep work — plan these. One per cycle, resourced properly.
- Medium impact, Quick win — good filler when the team has capacity.
- Low impact, Deep work — dismiss without guilt.
Filtering to Alert first thing on Monday is a reasonable habit; those are the items where something changed rather than something is merely improvable.
Dismissing
Dismiss a recommendation you have decided against and it leaves the queue. Dismissed items are kept in their own list rather than deleted, for two reasons: someone will ask why in a month, and a recommendation dismissed as low priority can age into a high one when the underlying gap widens.
Review the dismissed list monthly. Dismissing is a decision, not a delete.
Where they come from
Recommendations are derived from the rest of the platform, not from generic best practice:
- Prompts you lose in Prompts, especially high-intent ones.
- Competitor gaps from Overview and Competitors.
- Domains and pages the engines rely on in Sources.
- Page defects from Pages.
- Fan-out queries with no good answer of yours in Query Fan-out.
Which is why a recommendation that looks generic usually is not — open the analysis and the specific prompts and answers behind it are named.
Closing the loop
After shipping a fix:
- Re-analyze the affected page in Pages.
- Give the engines a cycle or two to re-crawl and re-answer.
- Check whether the target prompts moved.
- Look for a Win card confirming it, and repeat the pattern elsewhere.
A recommendation you shipped but never measured taught you nothing.