Skip to main content

Report labs: write it, grade it, schedule it

Three labs put the rest of this chapter into practice: write the page before you look at the numbers, grade a report you did not write, and fix the cadence while nothing is riding on it.

Labs​

Lab 26.1 — Write the page before you generate the evidence​

Lab · Where: Overview → Reports (/b/{businessId}/overview) · Cost: paid · Time: ~30 min

You need: your frozen baseline (Lab 7.3), a re-measurement (Lab 17.1 or 17.2), and the change log from Lab 17.4.

  1. Open a blank file. From your change log and your memory, write the four answers in plain language, one page maximum. Write no numbers yet — put [?] where a number belongs.
  2. Fill each [?] from stored data only: the keyword position chart, the grid trend line, Profile score over time, Review momentum. Write each number's conditions beside it. Any [?] you cannot fill this way is a claim you cannot currently support — leave it marked.
  3. Open Reports in the overview page header and press Generate. The price is on the button. The row moves through Queued and Generating; download it when it reads Ready.
  4. For every claim on your page, find its support in the annexe or a stored screen you can screenshot. Anything unsupported gets deleted, not softened.
  5. Mark every surviving movement V, P or U, writing each P's caveat in the words you would say out loud. Save the page and the dated PDF into one dated folder.

What good looks like. A page a non-specialist could act on alone, and an annexe where any claim on it is locatable in under a minute. Deleting at least one claim at step 4 is the normal outcome and the point of the exercise.

If it went wrong.

  • The row reads Failed. Generation is deterministic and is not retried, because a retry would fail identically. The badge is all the menu shows — the reason is recorded against the report rather than rendered beside it — and a failed run is refunded automatically.
  • A row is tagged Free analysis. A report generated for this place before the account existed. A valid dated artefact, but not one you produced this period.
  • You have no verified items. You measured a period you did not work in. Say exactly that; "no changes shipped this period, here is why" is survivable, invented activity is not.
  • The performance figures do not match your reporting period. They will not — that section uses a fixed window of its own. Quote it with the window named, or use the Performance panel instead.

What you just learned. Writing the narrative before looking at the numbers inverts the usual order, and that is why it works: you cannot cherry-pick a headline metric if you wrote the headline before you saw the metrics.

Lab 26.2 — Grade a report against the rubric​

Lab · Where: your own files (no SEOG) · Cost: free · Time: ~20 min

You need: a local SEO report you did not write — one a prospect shows you, a published sample, or your Lab 26.1 page graded by somebody else.

  1. Score the six rubric checks, pass or fail, no partial credit.
  2. Go number by number and mark whether each states its conditions. Write the ratio down.
  3. Find the tracked keyword set. If it is not stated anywhere, that is a fail on check 2 whatever the set contains.
  4. If you have two consecutive periods, diff the "next steps" sections.
  5. Write the single question you would ask its author that they would least like to answer.

What good looks like. A score out of six and one sharp question. Note that a low score is not proof of bad work. It is proof the work is unverifiable, which from the client's chair is the same thing until the day it isn't.

If it went wrong. It is entirely screenshots. Grade it anyway; "unverifiable" is a grade, and it is the grade. You wrote it yourself and scored six. You are not the right grader — hand it to someone with no stake in it.

What you just learned. The rubric is cheap to apply and expensive to pass. That asymmetry is what makes handing it to your own clients worth doing.

Lab 26.3 — Fix the cadence before you have a reason not to​

Lab · Where: your notes, plus free screens in the app · Cost: free · Time: ~15 min

You need: Lab 26.1.

  1. Write two recurring dates: the monthly work-report date and the quarterly (or fortnightly) measurement date. Put them in a calendar now, while nothing is at stake.
  2. Freeze the tracked set: every keyword row with its text, language chip, location chip, and the grid preset you will always use for it. It is an appendix in every report; changes to it are dated lines.
  3. Write, in advance, the sentence you will send when a measurement lands badly. One sentence. This is the whole lab.
  4. Set the archive rule: ten reports are kept per business and an eleventh prunes the oldest, so copy each PDF into your own storage on generation day — starting with the baseline, the first one you lose.
  5. Write the standing caveat that appears in every report: what you do not measure, and why.

What good looks like. Dates in a calendar, a written frozen set, an archive rule, and a bad-news sentence drafted before there is any bad news.

What you just learned. Every failure mode in this chapter is a decision made under pressure on reporting day. Making those decisions in advance, in writing, when nothing rides on them, is the whole defence.

Common mistakes​

Generating the report first and writing the story around it. The order determines the story, and reversing it is free.

Leading with the profile score. It is a diagnostic input, and a completeness audit that cannot see distance (Diagnosing a business in thirty minutes). Put it on the front page and the client starts optimising it — ticking low-weight attribute boxes — and you taught them to.

Reporting inputs under the outcomes heading. Rarely a lie in intent. Always a lie in effect, because the reader takes a heading at its word.

Smoothing a bad month. A client who later discovers it does not lose confidence in that month. They lose it in every month, including the true ones — and they are right to.

Shipping an unsigned report. Automated assembly is fine. Automated judgement with nobody's name on it is not a report, it is a printout.

Check yourself​

  1. Pick any number in your most recent report and state its conditions. Surface, query, coordinate, date — or window, or preset. If you cannot, either the conditions go in or the number comes out.
  2. Your tracked set changed this period. Where does the client learn that, in whose words? If the answer is "they would have to compare two PDFs", it is not disclosed.
  3. A client asks what they got for this month's fee, and every movement is P or U. What do you send? (Question 1 in full, question 4 with reasons, the honest sentence about the rest.)
  4. You inherit last quarter's report from a previous supplier. Which two rubric checks do you run first, and why those? (What you inherit with a client covers the rest of the handover.)
  5. Which claims in your last report would survive a successor paid to find the flaw? If the answer is "all of them", you have not tried it.

Next: The ninety-day plan →