What Google changed and how to measure it: the small-business guide

Between 18 and 27 August 2026, Google changed three things: how it ranks websites, how it reads their code, and the path a visitor travels between the search results and your page. None of these changes appeared on any screen. None sent a notification. Yet each one can move your numbers, or make them look as if they moved, which is not the same thing. This guide explains how to tell the difference, with a method that fits into four steps and needs no paid tools.

Three changes in ten days, and why nobody saw them

On 18 August, Google rolled out an update to its spam system, completed on the 21st, worldwide, in every language. The company did not specify what the update targets, and said nothing about AI-generated content, contrary to what many commentators claimed. After this kind of update, some sites rise, some fall, and panicked owners rewrite pages that asked for nothing. If your traffic has moved since, start with our five-check protocol: Your Google traffic dropped: five checks before you touch anything.

On 21 August, a quieter change still: Google stopped fixing certain encoding errors in structured data on your behalf, the invisible block of code that tells it "this is a price, this is a rating, this is the company name". The error triggers no alert, the validation test can stay green, and the recorded value changes in silence. The full verification walkthrough is here: Structured data: why the green light is no longer enough.

On 26 August, Google confirmed the rollout of a redirect between its results and websites: a visitor who clicks now passes through an intermediate Google address before reaching you. Its effect on your analytics: unknown at the time of writing, and that uncertainty is managed precisely by measuring. The details: google.com/goto: the invisible door between Google and your site.

Three different natures: a ranking, a code reading, a click path. One thing in common: none of the three can be repaired blind. That common trait is what justifies a guide.

The expensive reflex: repairing before measuring

When numbers fall, instinct pushes you to act. Rewrite pages, change providers, accept the cold call promising to "recover your rankings" for a subscription. That instinct is wrong twice over. First because a late-August drop can come from the update, but also from the end of the holidays, from your customers' own calendar, or from a change you made to the site yourself: two bad weeks prove nothing. Second because a repair launched without a diagnosis destroys the one thing that would have allowed you to understand: the reference state. If you rewrite ten pages while rankings are moving, nobody will ever know what produced what.

My previous profession, scientific research, has a name for this discipline: controlling the variables. You fix a baseline, you change one thing at a time, you never conclude from a single data point. None of it requires a laboratory. A notebook is enough.

The method: four steps, no paid tools

1. Set your baseline, today

A baseline is the state of your numbers before anything is touched, dated and kept outside the tool that displays them. Without it, "it dropped" and "it's recovering" are moods. With it, they are measurements. Three series are enough for a small-business website: your clicks in Search Console, your "google / organic" sessions in GA4 or whichever analytics tool you use, and your share of direct traffic. Note all three, with the date. A dated screenshot does the job perfectly.

2. Keep the change log

The notebook's second column: what changed, and when. Your own changes first, because they explain more traffic drops than Google does: a redesign, a migration, ten deleted pages, a new theme, a caching plugin. Then dated external events: an update announced on Google's official status dashboard, a press article that mentions you, an advertising campaign. When a curve moves, this log turns the question "why?" into a simple lookup: what happened that week, at your end or outside?

3. Read the numbers together, never separately

This is the step almost nobody takes, and it is the most powerful of the four. Each tool measures something different: Search Console counts clicks on Google's side, your analytics tool counts arrivals on your site's side. As long as both tell the same story, the variation is probably real. When they diverge, you are holding a signal of a different nature.

A concrete example with August's redirect: if one day Search Console stays flat while your analytics loses "google / organic" traffic to "direct", demand has not fallen, the origin label has shifted. That is not yet a cause, it is a signal worth investigating, and it is already far more than a shiver. The reverse is instructive too: clicks falling in Search Console at equal positions, while your pages stay well ranked, point to a change in how results are displayed rather than to a problem with your site.

4. Decide on four weeks, not on two days

Google's documentation is explicit: its spam systems are re-evaluated periodically, without an announcement each time. Rankings can therefore move again after an update, in either direction. Add seasonality, the back-to-school period, the weather of your own trade, and the conclusion writes itself: a decision made on two days of data is a bet, a decision made on four weeks is an informed choice. Set in advance the date on which you will reopen the notebook, and keep it. That future appointment is also your best protection against the urge to touch everything in between.

When to act immediately, all the same

Measuring first does not mean never doing anything. Three situations justify immediate action, because the diagnosis there is certain. A manual action shown in Search Console: that is not an algorithm effect, it is a notified sanction, and it gets handled. A wrong value in your structured data, confirmed in the official test: your company name or a price recorded incorrectly will not fix itself, and the check takes two minutes. A compromised site, pages added without your knowledge or redirects you never created: every day counts. In every other case, measurement comes first.

What this guide does not claim

Nobody knows the detail of Google's algorithm, and this guide does not pretend to. Nor does it promise that good measurement lifts a site's rankings: it does something better, it tells you whether the problem exists, where it lives, and whether the money you are about to spend has a target. The first published figures on AI Overviews in France illustrate the nuance: the Ahrefs study of August 2026, run on 963 high-traffic domains, measures a 23.1% drop in click-through rate for the most exposed sites, over a nine-day window, at the height of summer, published by an SEO tool vendor. That is a serious first signal, not a verdict, and above all it is not YOUR number: your own exposure takes ten minutes to check with your five vital queries. The general measurement informs; only yours decides.

Three questions close this guide, the same three that open every diagnosis we run: what do your numbers say, since when, and compared to what? The day those three answers fit on one page of a notebook, nobody can sell you a panic again.

Frequently asked questions

More questions? Many answers are already gathered on our FAQ page.

Information current as of 27 August 2026; the dates and rollout status of the changes described may evolve. Main sources: Google Search Status Dashboard, spam update of 18 to 21 August 2026 · Google Search Central announcement of 21 August 2026 on structured data processing · Google's confirmation of the click redirect, 26 August 2026 · Ahrefs, study of the click drop linked to AI Overviews in France, August 2026, 963 domains, aggregated Search Console data. Cover photo: George Milton, Pexels.