Automate a Weekly GEO Brief with Claude and Qwairy MCP

A weekly brief is only worth building if it runs without you. This one is a Claude scheduled task: the same tools in the same order on the same weekday, ending with three actions posted to Slack and logged in Notion.

The GEO problem it solves

The failure here is cadence and delivery, not analysis. A dashboard opened when someone remembers produces numbers with nothing to compare against, and a brief with no owner produces no work. The schedule fixes the first, Slack and Notion the second.

One reading trap is worth naming. A headline number hides which component moved, and mention rate and source rate can diverge in the same week. The overview returns a numerator and a denominator for three of its four scores, which separates a real shift from a thin week. Sentiment has none.

How the workflow breaks down

1. Schedule it. Create a Claude scheduled task on a fixed weekday and hour, with the brand and the period written into the prompt itself. Anything decided at run time makes consecutive briefs incomparable.

2. Measure the change. get_overview with period 7 returns the four scores next to their previous-period values, with numerators and denominators for three. A denominator of zero means no monitoring ran: the brief has nothing to compare and must say so rather than report a drop to zero. If source rate and mention rate diverge, get_page_presence gives the cited-page snapshot.

3. Locate the movement. get_matrix crosses topics with providers. get_competitor_comparison is already restricted to you and your direct rivals; get_competitors widens that to every tracked relationship, on all-time counts. get_competitor_evolution returns daily mentions, position and sentiment one competitor at a time, and get_competitor_position adds coverage, share of voice and top citing domains for a mover.

4. Reduce to three actions. get_content_opportunities returns up to ten missed questions where direct competitors are named without you, plus questions where you appear in under 30 percent of answers. get_prompt_signals returns the four action counts and the two shortlists. Neither estimates revenue.

5. Deliver, then record. Post to a fixed Slack channel: the metrics with their denominators, what moved, then the three actions. Create one Notion item per action with the metric it targets, its value today, an owner and a review date. If either is not connected, return the brief in the chat and name what failed.

What a run actually returns

A real week: mention rate 1.08 percent on 10 of 928 responses, flat. Source rate 3.49 percent on 54 of 1548, up 1.2 points. Share of voice 0.25 percent, flat. Sentiment 60.5.

One component moved and three did not, which is the ordinary case and why a headline number is useless weekly. The brief opens on source rate, the only line with a change to explain, and names its denominator so next week can tell a real move from a thin sampling week. The three flat lines still get printed: hiding them makes the week something does move unreadable.

What to do with the output

Rank the candidates in this fixed order and take the top three: a question appearing both as a low-visibility question and in the attack shortlist, then missed questions naming two or more direct competitors, then everything else. Break ties on more answers behind the question, and publish fewer than three if fewer exist.

If content opportunities comes back empty and the signals put everything in ignore, the honest brief is one line saying no action qualified, plus the metrics. Post that line: a week with no action is information, an invented one is not.

The Notion item is your baseline: it carries the metric value on its creation day, which is what next week's run compares against, while the tool's previous period moves with the window.

The Complete Prompt

Give me my weekly GEO brief. List the brands I monitor and ask which to use if there is more than one. Ask once for the Slack channel and Notion destination, then reuse both weekly. State what each step shows before moving on.

1. Run get_overview with period 7. Compare mention rate, source rate, share of voice and sentiment against the previous 7 days, quoting numerators and denominators. If a denominator is zero, say no monitoring ran, post that to Slack and skip the rest. Never lead with sentiment, which has none. If source rate and mention rate diverged, run get_page_presence.
2. Run get_matrix for the last 7 days. Show the topic by provider grid and name the cells where one provider clearly lags. Treat gaps as places to look, not diagnoses.
3. Run get_competitor_comparison, which already covers me and my direct rivals only. Run get_competitor_evolution over 14 days for the top 3, reporting daily mentions, average position and sentiment only. For each competitor gaining mentions, run get_competitor_position and present its citing domains as a map, not as proof of cause.
4. Run get_content_opportunities and get_prompt_signals. Rank candidates in this order: questions both low-visibility and in the attack shortlist, then missed questions naming two or more direct competitors, then the rest. Break ties on more answers behind the question and take the top three. If nothing qualifies, say so and continue to step 5 anyway.
5. Post the brief to Slack: the metrics with their denominators, what moved, then the actions, each quoting the metric behind it and labelling hypotheses as such. Then create one Notion item per action with the target metric, its value today, a proposed owner and a review date one week out. If Slack or Notion is unavailable, print everything here and name the destination that failed.

Do not invent metrics, competitor names, traffic estimates or a time to first citation.