Package GEO as a Repeatable Agency Service

Selling GEO one client at a time works until the third, when every audit and report gets rebuilt from scratch. This workflow standardises two artefacts: the audit a prospect sees before signing, and the report a client gets every month after.

The GEO problem it solves

Agencies tend to package GEO like an SEO retainer: a dashboard, a deliverable list, a price anchored on the retainer next door. That breaks immediately: there is no shared ranking table to point at, so the prospect asks what they are buying and gets an activity log.

Both artefacts already exist as data. A Pitch Audit is a one-off run against a prospect domain, scored before any engagement exists; a monitored brand carries the recurring measurement afterwards. Run in that order, what you sell against and what you report against are the same measurement.

How the workflow breaks down

1. List the pitch audits. list_pitch_audits returns the prospect audits in the agency workspaces your token can reach: name and domain, status, market, models, prompt and competitor counts and score summary, 20 per page up to 100. An empty list, or an unavailable tool, means no agency workspace or no audit yet: go to step 3 and baseline a monitored brand.

2. Open one audit. get_pitch_audit returns the report for an audit ID: status and current stage, prospect profile, prompts, competitors, and a summary carrying the scores, the competitor leaderboard, the sources and the recommendations. A subject ID re-renders it from a competitor's point of view. On a run that has not produced a report the summary comes back null: unfinished, not a prospect without competitors.

3. Freeze the contract baseline. Once the client is monitored, get_overview returns mention rate, source rate and share of voice, each with numerator and denominator, plus sentiment, the previous window and the last monitoring date. Store it, dated, before any optimization work starts.

4. Build the monthly report. get_visibility_trend returns the series plus a direction the tool sets itself: up above two points of visibility rate between the window halves, down below minus two, stable in between. get_content_opportunities returns up to 10 prompts where competitors are mentioned and the client is not, up to 10 under 30 percent visibility across at least two answers, and a flag tripping above 5 missed or 3 low-visibility prompts. get_technical_status adds the readiness score, blocked crawlers and open issues by severity.

What to do with the output

Rank the portfolio once per cycle, before allocating analyst hours. Clients whose trend covers zero answers go first: on an empty window the direction still reads stable, so an unmeasured client looks exactly like a healthy one. Then clients whose direction is down. Then clients whose opportunity flag is raised, ordered by the lowest visibility rate among their returned low-visibility prompts. Everyone below that line gets the report and none of the hours.

Send the report either way, minus two claims: a delta against a previous window with no answers behind it is not a gain, and a positive delta is not proof the agency caused it. Give the movement, the window and the denominator.

The Complete Prompt

I run GEO for client brands and want the same two artefacts every time: the audit I show a prospect, and the report I send a client. List my pitch audits and my monitored brands, then ask which one to work on.

1. List my pitch audits with status, prospect domain, market, models, counts and scores. If the tool is unavailable or returns nothing, say so and go to step 3 with a monitored brand.
2. Open the audit I pick: status, profile, competitors, and the summary with its scores, sources and recommendations. If the summary comes back null, say the run has produced no report and stop, rather than reading an empty competitor list as no competitors. Offer to re-render it from a competitor's point of view.
3. For a monitored client, pull the overview over the last 30 days as the contract baseline: mention rate, source rate and share of voice with their numerators and denominators, then sentiment, the previous window and the last monitoring date. Flag it when that previous window has no answers behind it.
4. Pull the visibility trend, content opportunities and technical status for the same window into the monthly report: series, the direction the tool returns, prompts where competitors appear and the client does not, prompts under 30 percent visibility, open issues by severity, last month's gaps. If the trend covers zero answers, report the client unmeasured, not stable, and say so when no technical analysis has run.
5. Rank my clients for the cycle: unmeasured windows, then a down direction, then a raised opportunity flag ordered by the lowest visibility rate among the returned low-visibility prompts, then everyone else. State which get analyst hours and which only get the report.

Never present a score change as proof my work caused it. Report the movement, the window and the denominator behind it.