Actions
Where recommendations come from, the bar they must clear, and how their status moves.
On this page
Actions turn what a diagnosis finds into work you can execute and sign off. PeakGEO follows one rule: verify first, then improve. An answer that looks weak does not automatically mean your brand is missing content. Improvement work is proposed only after a problem is confirmed, and its effect is observed by retesting the same prompts.
Where actions come from
Evidence gaps create verification actions
After each diagnosis, every gap in the evidence chain becomes a verification action. No citation, an unreachable source, a citation that doesn't support the claim, verification still pending, and genuinely missing content are different situations, so a verification action asks you to confirm the facts. It never tells you to build a new page on its own.
Verification actions keep the gap's own ownership. They don't invent an owner, difficulty or timeline, and different gaps are not merged just because they would produce a similar deliverable.
Material gaps create material actions
When a gap is about missing brand material (for example, product or use-case details the prompts need), you can create a material action straight from the gap. If an open action already exists for that gap, it is reused rather than duplicated.
If some prompts couldn't be written because the brand info isn't specific enough, the console asks you to add brand info and run the diagnosis again. No improvement actions are created from that run in the meantime.
Verification findings lead to improvement actions
Once you record a finding in the action's Verification and findings tab, use Analyze next step, or select actions in the list and choose Generate action plan. The analysis only uses existing answers and materials. Nothing is retested or executed automatically.
The bar for an improvement action
An improvement action needs a clear basis. It must meet at least one of these:
- the verification it follows up on is marked Issue found; or
- its linked evidence shows the AI answer is unsupported or conflicts with the facts.
Evidence showing that a source supports an answer means the answer is fine, so it can't justify an improvement.
Every proposal is validated individually
Each action the model proposes is checked on its own. Only those that pass go to To do. Common reasons a proposal is rejected:
| Reason | What it means |
|---|---|
| No basis | No verification marked "Issue found" and no evidence that the AI got something wrong |
| Issue not confirmed | The verification it follows hasn't been marked "Issue found" yet |
| Missing fields | Missing the task, reason, deliverable or acceptance criteria |
| Promised results | Acceptance criteria promise an outcome such as a higher mention rate |
| Prompt not measured | A linked prompt wasn't tested this time or isn't in the formal prompt library |
| Evidence mismatch | Cited evidence is no longer available or doesn't match the linked prompts |
| Duplicate | Repeats an existing action or another proposal in the same batch |
Rejected proposals appear under Proposals not adopted with their reasons, so you can decide whether to add material and analyze again.
What an action contains
Each action states the task, why it matters, the deliverable and the acceptance criteria, and links to specific prompts and evidence. Improvement actions also show the retest scope (the linked prompts) and the retest metric. Acceptance criteria measure whether the work was done as required, not whether a metric went up.
There are no priority tiers. The list order is the suggested working order: verification actions follow the order of their gaps, and improvements derived from a verification come right after it.
Status flow
| Status | Meaning |
|---|---|
| To do | Created, not started |
| In progress | You chose Start verification or Start execution |
| Awaiting retest | The improvement's completion record is saved; waiting for a retest |
| Retested | A retest result is linked to the action |
| Closed | The action is finished |
Verification actions
To do → In progress → Closed. When you save the verification record, you add notes and choose a finding:
- Issue found: the verification closes, and you can run Analyze next step to propose improvements.
- No issue found: the verification closes; nothing needs improving.
- Insufficient material: the action stays in progress, and you note what is missing and which role should supply it.
Verification actions don't schedule an effect retest.
Improvement actions
To do → In progress → Awaiting retest → Retested → Closed. Complete the work against the deliverable, then save a Work completion record (notes plus links to deliverables). The action moves to Awaiting retest; saving the record doesn't start a retest by itself. If the work needs new content, choose New topic on the action to continue in Content.
An improvement action can only be closed after it has a retest result. Work completion and outcome are assessed separately: without a retest, the action can't be marked Closed.
You can start a retest manually from Monitoring & retests, and one is scheduled automatically when you add a published link for content. Retests only compare results gathered under matching conditions; see Monitoring.
What happens after a new diagnosis
Baseline verification actions refresh with the latest gaps:
- not-yet-started verifications whose gap still exists are updated with the current task, rationale and linked prompts;
- not-yet-started verifications whose gap has disappeared are removed;
- actions you have already started, and improvements proposed by analysis, stay as they are, so your work records are never lost.
A partial retest does not change the action list.
Interpretation limits
- Actions don't promise that AI will recommend your brand, or that rankings or mention rates will rise.
- A change in a single retest can't be attributed to one action alone; model versions, prompt scope and sampling conditions all affect results.
- The diagnosis report lists improvement actions only; see Report export.