Algorithm update week: triage before narrative

Update week runs on a script, and you can watch it run. Your metric wobbles on Monday. By Tuesday the industry chatter is in a channel. By Wednesday there is a meeting called something like update response, eleven people are invited, and the agenda already assumes an answer.

Nobody on that invite has checked whether the wobble is real.

Triage before narrative is the replacement script. Three questions, in order. Is the change real. Is it you or the whole results page. Where does it concentrate. Only then does a recovery conversation earn its meeting. The order matters because each answer changes who owns the next step.

First, prove the change is real

Rollout weeks are volatile for everybody. That is why the test goes first.

During a rollout, ordinary variance reads as update damage more often than in any other week of the year. The base rate of coincidence peaks exactly when the room's tolerance for it bottoms out.

So run the significance test on your delta before you treat it as an event. That is a statistical check on whether the change is bigger than your metric's normal week-to-week wobble.

The player below shows that read mid-rollout: the movement, the test, and an answer separating what the update did from what the week's weather did.

A real result opens the investigation. A noise result closes it. Closing an investigation during update week is a genuine save, because that is the week everyone is most tempted to act on nothing.

What did the page itself become?

Often something other than a reshuffle of who ranks where.

The sharper update-week read watches the whole surface rather than your own row. During one rollout, a team's weekly check reported movement across their entire page-one set, momentum switching between competitors on non-branded terms, and more AI Overviews appearing, meaning the AI-written blocks Google shows on top of a classic results page.

Citation activity rose with them, which is how often an AI answer names a page as a source out of the questions being tracked. That team's own line was one row in a much larger picture.

The framing changes what you are asking. Did the update reshuffle rankings, or did it change what the page shows: more AI Overview real estate, different features, different winners?

A drop caused by the page changing shape needs different work from a drop caused by your pages losing rank. Only the wide view tells the two apart, and the competitor and feature series over time make it a reading.

When waiting is the right call

When it arrives with a written list of what you would check next.

The hardest moment of update week is organisational. A leader whose weekly number wobbled mid-rollout gets asked about it in the executive meeting, and the pressure to announce an action is real even when acting is wrong. We have watched a growth leader hold that line through a rollout and keep the room. The position came with structure instead of hope.

The structure was a triage list: every candidate explanation written down, each with the check that would confirm it or kill it.

• Seasonality, read against the same weeks a year earlier.
• Click-through shifts caused by the page changing shape.
• Paid and organic interplay moving the mix.
• Pages you removed or redirected inside the same window.
• Tracking composition, where the set you measure changed rather than the world.

The baseline does most of the work

Triage goes fastest for the teams that were already watching.

A wobble with a hypothesis list attached reads as control. A wobble without one reads as drift, and drift gets a reorganisation nobody needed. Rollouts also take time to settle, which makes an early reaction doubly expensive. You change course on a half-finished redistribution, then change again when it finishes.

A standing weekly check means your pre-update baseline exists as a series instead of a memory. The significance test then has real variance to work with, rather than two cherry-picked weeks. Update week is that same weekly routine with the volume turned up, which is why teams who already run it finish triage on Tuesday.

The baseline also settles the most common argument of the week. Was the level you fell from itself unusual? A spike that settles reads as a loss to anyone who joined the chart late. A longer series puts it back in context, and the weekly reporting article covers the routine that builds one.

See also: the weekly routine that builds this baseline before you need it →

Where does the change live?

In one content family, one intent cluster, or one template, nearly always.

A real, update-shaped change still is not actionable until you know where it sits. Break it down by category, intent, device and page family, and the concentration is usually sharp, computed on the same pages across both periods so the comparison holds up.

The worked example below shows that read in motion, a category-level decline arriving with its statistical result attached. That is the artifact an update week deserves. Located, tested, and ready to hand to whoever owns the category.

The workflow that does this: Category decline, verdict included →

Recovery is the work you already had

Updates rarely invent new work. They reorder the work you already had.

The post-triage list is your standing one, reordered: refresh the decayed pages in the category that got hit, fix the technical faults the update exposed, strengthen internal links to the pages that slipped. What triage adds is aim, plus confidence that the aim is real and located.

Then report it the way it happened. The update-week slide carries the result, the location and the as-of date, meaning the day the data was last complete, printed so a lag is never mistaken for a loss. What you leave behind shows a team that measured first and moved second.

Two cycles of that and update week stops being frightening. It becomes a week with a slightly busier Tuesday.