Running the week

What keeping the clients healthy actually looks like, day to day.

This is a first draft. It was written from how the platform behaves, not from having done the job for a year. Correct it as you go — especially the timings, which are guesses about how long things take.

#The platform's own clock

Everything below hangs off what the system does automatically.

The schedule is set in UTC. The UK is on GMT in winter (same as UTC) but BST in summer (UTC+1), roughly late March to late October. So for about seven months of the year everything below happens an hour later than the raw schedule says.

So: Monday morning is the week. By 06:30 UTC — 7:30am while we're on BST — the platform has re-scanned everything, scored the changes it made a month ago, and produced this week's recommendations, Google post suggestions and blog topics. Everything else is lighter.

If you start before that, you're reading last week's data. Worth knowing in summer, when the gap is an hour wider than you'd expect.

#Every morning (20–30 minutes)

#1. Declines first, always

Detection ran at 05:30. A ranking that has dropped is more urgent than a ranking you could gain — a page that fell off page one is losing money today, and the cause is usually something we did.

For each new decline: look at what changed on that page recently. The platform correlates changes to declines, so start there. If one of our changes preceded the drop, say so and roll it back — don't defend it.

#2. The leads inbox

Every client, every new enquiry. Two things to check:

• Did the client actually receive it? A lead in the dashboard that never reached their inbox is worse than no lead, because everyone assumes it was handled.
• Does it look real? Spam gets through. Don't delete it — flag it. Deleting leads is admin-only for a reason.

#3. Health probe

The 07:00 probe flags sites that are down, slow, or serving errors. Anything red gets looked at before anything else on this list. A client site being down is the only true emergency here.

#Monday (2–3 hours)

#1. Read the verdicts before the recommendations

The platform scores changes made ~28 days ago as better, neutral or worse against a real baseline. Read these first, because they tell you whether last month's judgement was any good.

A run of "worse" verdicts on one kind of change means stop making that kind of change. This is the only honest feedback loop in the job — use it before deciding this week's work.

#2. Work the recommendation queue

New recommendations landed at 06:00. For each one:

• Does the target URL exist? Open it. The AI has invented plausible-looking URLs that 404 — six out of ten on one report. Never approve a change pointing at a page that isn't there.
• What does this page already rank for? If the change removes a phrase the page ranks on, that is a ranking risk. The platform flags it and won't let a non-admin approve it. Rewording to keep the phrase is almost always better than escalating.
• Is the claim true? Anything asserting a review count, an accreditation, insurance, years in business or a case study is rejected unless the client has confirmed it. On fire-safety and trade sites this is absolute.
• Would you say this to the client's face? If the copy is generic filler, reject it. Thin content that reads like AI does not rank and does not convert.

Reject freely. A rejected recommendation costs nothing; a bad approved one costs a ranking.

#3. Rank tracking sanity check

Don't trust the tracker blindly — it has been wrong. It once reported a page at #1 that was invisible in a real search, because it wasn't looking deep enough into the results.

Once a week, take two or three keywords per client and cross-check the tracked position against Search Console's average position for the same query. If they disagree by more than a few places, the tracker is wrong, not Search Console. Flag it rather than reporting the flattering number.

National rankings are reliable. Local ("near me", town-level) positions from the tracker are not — they come from a datacentre, not from the town. Don't put local positions in front of a client without checking.

#4. Google Business Profile posts

Paste this to do it with Claude:

Weekly Google Business Profile round for every client. For each site with a Business Profile
connected: tell me whether a post is currently live or the profile has gone quiet, generate
suggestions if none are waiting, and show me each one in full. For every suggestion flag
anything the client may not be able to evidence — certifications, accreditations, insurance,
review counts, years in business, job numbers, response times, guarantees — and check the
button link points at a real page. I'll decide what publishes.

Do this every week without fail. It is the shortest task on this list and, for a local trade, usually the highest return per minute spent.

Two facts drive it:

• A post disappears after seven days. Not archived — hidden. A profile with no live post looks less active than a competitor's, in the one place people are choosing who to phone.
• It shows in the map pack, where a small business can out-rank a much bigger one.

Per client, on the site's Google posts page:

1. Press Suggest posts if nothing is waiting. Two arrive, drafted from that client's own services, areas and recent market-intelligence findings.
2. Read each one properly. It publishes word for word, in the client's name, and someone may ring the business because of it.
3. Publish one or two. Reject the rest — an unpublished suggestion costs nothing.

What you are reading for, and this is the whole reason a person is in the loop:

• Any claim the client cannot evidence — a certification, an accreditation, insurance, a review count, years in business, number of jobs done. The generator is instructed never to write these. Check anyway. On a fire-safety or electrical client an unevidenced claim is a legal problem, not a tone problem.
• Anything that reads as a promise: response times, guarantees, prices.
• The button link. It is validated against real pages, so it should never 404 — but check it goes somewhere sensible for what the post says.

If a published post turns out to be wrong, Retract from Google takes it down. That is the only way to withdraw one, so use it rather than leaving it for seven days.

Cadence: aim for one or two a week per client. Rejecting everything one week is fine. Publishing something you haven't read is not.

#5. Blog and content schedule

Topics now arrive on their own — the platform suggests them Monday morning, so this is a review job rather than something you have to remember to start.

Check what's queued, what's drafted and what's published. Two rules:

• Nothing publishes without being read end to end by a person.
• A post exists to answer a question people are actually searching. If you can't name the query, it isn't ready.

Keep a steady cadence rather than bursts. The platform limits how much changes per week on purpose — over-tinkering makes the verdict scoring meaningless because you can't tell what caused what.

#6. Keyword and area coverage

Per client, ask: are we tracking the words they'd actually be hired for, and the towns they actually serve? New services and new areas come up in conversation and never make it into the platform.

Adding a keyword or an area is member-level work — do it.

#Monthly

• 1st: AI visibility runs. Worth a look, don't over-read it yet.
• Reports: each client's monthly report should be checked before they see it. Numbers come from Google and their own site — if something looks impossible, it probably is.
• Business Profile data lags two days by design. A "missing" recent day is not a bug.

#Things that are not your call

Escalate rather than decide:

• Approving a change the platform flags as risking a live ranking
• Deleting a lead
• Connecting or changing a Google Ads account
• Anything involving DNS or a client's email
• A client asking for something that would mean stating a claim we can't evidence

#The judgement, in one line

Most of what goes wrong is one of three things: changing a page that already ranks, believing a number without checking it against Search Console, or publishing something we can't evidence.

If you catch yourself about to do any of those, stop and ask.

#What "running smoothly" actually means

At the end of a good week: no client site is down, every enquiry reached its client, every decline has either been explained or reversed, this week's recommendations have been worked through rather than left, and nothing has been published that anyone would be embarrassed to defend.

Not: "the queue is empty."

Internal documentation. Not for clients, and not for search engines — this page sends noindex, but the link itself is the only thing keeping it private.