Measurement Loop — Peec AI → Content Backlog

This is what turns the engine from an order-taker into a visibility strategist: write what moves the client's Share of Model, not just what someone happened to ask for. Reads brand-visibility data from the Peec AI MCP; tracks it in the client's Target Prompts & Visibility sheet.

The metrics that matter (2026)

Traditional rank/CTR is replaced by: - Share of Model / Share of Voice — how often the client appears in AI answers vs. competitors for the category's prompts. - Mention rate & citation frequency — appearances and linked citations. - Sentiment — how positively the brand is portrayed. - Citation provenance — which domains drive the client's inclusion.

Why one-off checks lie

AI answers are probabilistic and volatile — studies show only ~30% of brands stay visible from one answer to the next, and ~20% across five runs. So visibility must be sampled repeatedly, not spot-checked. The credible method: define a fixed set of 250–500 high-intent prompts for the client/category and run them on a regular cadence to get statistically stable estimates.

Target Prompts & Visibility sheet

Created at onboarding (socrates_upsert_target_prompts). Columns: prompt, intent stage, client visible? (y/n), competitor(s) cited, sentiment, last checked, target content (the piece that will address it), status. This is the client's prompt "population proxy" and the running scoreboard.

The loop

1. Pull the client's prompt set and current standing from Peec AI.
2. Find the gaps — prompts where the client is absent or a competitor wins. These are the highest-value topics.
3. Prioritise by intent and gap size into a content backlog.
4. Write the piece through the normal three-stage workflow, aimed squarely at the gap prompt(s) as target AI-prompts.
5. Re-measure on the next cadence; update the sheet; note movement against the Run Log so output can be correlated with visibility change over time.

Backlog mode

When asked to "find what to write for [client]" / "what are we losing in AI search," run steps 1–3 and return a prioritised backlog (prompt → gap → suggested piece/format), then offer to write the top item. Don't auto-write the whole backlog.

Slack alerts (optional, high-value)

Flood already runs Slack. Worth wiring Peec/monitoring to alert the client channel when sentiment drops or an inaccuracy/visibility cliff appears, so the team reacts fast instead of finding out at the next review. This is a monitoring-side setup, not a skill behaviour, but flag it as a recommendation during onboarding.

Honest limits

Visibility is not engagement — being cited doesn't guarantee clicks or conversions, and attribution from AI answers is still coarse. Track it as a leading indicator of brand presence, not a revenue number.