An unconfirmed Google update in September 2026 cannot be diagnosed from a tracker screenshot, and any article telling you exactly what Google changed is guessing. Nobody outside Google knows the mechanism. What you can know, in an afternoon, is whether your own site actually moved, which part of it moved, and whether the cause was Google at all.

That is the job. Not naming the update: measuring your site.

What an unconfirmed Google update actually is

Google confirms a small number of ranking changes each year, mainly core updates and spam updates, and publishes them on its Search Status Dashboard with a start date and an end date. Everything else ships quietly. Google has said repeatedly that it makes thousands of changes to Search a year, and it does not narrate them. So when rankings wobble and the dashboard is silent, third-party volatility trackers and forum threads fill the vacuum with inference.

The practical difference matters. A confirmed core update comes with documented guidance, a stated rollout window, and a shared timeline everyone can compare notes against. Unconfirmed volatility comes with none of that: no window, no guidance, no rollback story, and no way to tell a broad change from a narrow test affecting one query class in one country.

Set your expectation now. You are not going to identify the update. You are going to work out whether your site moved and why, and those are different questions with different answers.

Rule out your own site before you blame the algorithm

More "algorithm hits"are self-inflicted than anyone admits. Before you touch a word of content, check what you shipped.

• A deployment, theme change or plugin update in the same week

• A robots.txt rule or a staging noindex tag carried live with a redesign

• A broken or newly self-referencing canonical after a URL structure change

• An expired SSL certificate, or a hosting migration that changed response times

• Duplicate structured data after an SEO plugin change

That last one is quietly responsible for a lot of misattributed drops. If Yoast is already emitting Article or FAQPage markup and a second source starts emitting its own, you get conflicting markup on the same URL, and rich results can disappear overnight. Traffic falls, the timing lands inside a volatility window by pure chance, and the update takes the blame for a schema conflict.

A worked example of the pattern: a recruitment firm sees a 60% traffic fall and assumes an algorithm hit, then finds a staging noindex tag went live with a Framer redesign on 9 September. Search Console's Page Indexing report shows the "Excluded by noindex"count climbing from that exact date. No update involved. Check Page Indexing and Crawl Stats against your own release calendar before anything else, because a crawl or indexing cliff that lines up with a deployment date is not a coincidence.

Then rule out demand. September is a genuine step change in UK search behaviour for B2B, education-adjacent and professional services sectors, because the summer lull ends and budgets restart. When you are testing whether the movement is seasonal demand rather than ranking change, compare the same weeks against September last year as well as the like-for-like windows below.

How to read your data so the drop tells you something

Search Console data lags by roughly two to three days. Most people open a default 28-day comparison that straddles the onset of the volatility, which smears the event across both windows and dilutes the signal into nothing. Then they conclude it was mild, or that it was worse than it was, depending on where the dates fell.

Use matched, complete weeks either side of the suspected start date. For a mid-September onset, compare 15 September to 12 October 2026 against 18 August to 14 September 2026: four complete weeks each, same days of the week, no partial weeks at either end. Filter to the UK. Filter to one folder, say /blog/, so you are reading one thing at a time.

Then split brand from non-brand. A drop in queries containing your company name is almost never an algorithm story, it is a PR, seasonality or paid-media story. A drop across many non-brand queries at once usually is an organic search story. Reading a blended total tells you neither.

Separate the three signals, because they mean different things:

• Impressions down: you have lost visibility, so you are appearing for fewer queries or on fewer pages of results.

• Position flat, clicks down: you are still ranking where you were, but fewer people are clicking. That is a results page or a title and snippet problem.

• Position down across a whole section: a content quality or intent match problem in that section.

Search Console averages hide shape. Add daily rank tracking on a small representative set, twenty to thirty keywords covering your money pages and your best-performing articles, so you can see whether the move was a cliff, a slide or a bounce. One monthly average tells you a number. Daily tracking tells you a story.

Segment before you conclude. Page type, folder, device, country. "The whole site dropped"is almost always "one folder on mobile in one market dropped"once you look properly.

Demotion or SERP change? Telling the two apart

Here is the combination people misread most often: stable average position, stable impressions, falling click-through rate on the same queries. That is not a demotion. That is a results page that has changed shape around you.

AI Overviews, expanded People Also Ask panels, video carousels and shopping blocks all push the first classic blue link further down the page without touching anyone's ranking position. You are still number four. Number four is just further from the top of the screen than it was.

Take the example of a 40-page UK services site: non-brand clicks fall sharply from mid-September, brand clicks hold flat, total impressions hold flat, and the average position for the top ten commercial queries barely moves. Nothing in that pattern says content demotion. It says the click-through rate collapsed because the page around you changed. The fix is titles, snippets and intent match, not a rewrite of forty pages.

The other thing to test is an intent swap, where Google reweights a query from informational to commercial or back again. Your guide has not got worse. The results page now rewards product pages, and a guide is the wrong format for what is being served.

So run a manual check on your five most valuable queries. Open each in an incognito window, note what the top ten actually contains now (format, page type, publisher type, how much of the screen is organic), and compare it honestly to your page. This kind of SERP analysis for content planning takes twenty minutes and settles arguments that tracker charts cannot.

The patterns unconfirmed volatility tends to expose

Volatility rarely creates weakness. It exposes it. When a site moves during an unattributed episode, the pages that fall hardest are usually the same four kinds.

Thin templated pages. The ones that exist to hold a keyword rather than answer anything: same structure, different town name, nothing in them a reader could not have guessed.

Generic AI-assisted writing. Most AI writing has a tell. It is competent, it is grammatical, and it sounds like nobody in particular: vague filler, flat sentence rhythm, phrasing that could describe any firm in the sector. Pages carrying first-hand specifics hold up better, and that is not a claim about detection, it is a claim about usefulness. If you want the detail on where AI content genuinely gets sites into trouble, we covered the real triggers behind spam update hits on AI content separately.

Cannibalisation dressed up as a drop. This is the most common self-inflicted "update"we see. Two of your own pages trade places for the same query. At page level it reads as a collapse. At site level, nothing happened. Checking existing pages and previously published articles before you write anything new prevents most of it, which is why a duplicate check belongs in the pipeline before drafting, not in the post-mortem afterwards.

Experience and authorship gaps. On money, health and advice topics especially, pages with no named author, no stated method and no evidence of anyone having actually done the thing are the fragile ones. Google's own self-assessment questions on helpful, reliable, people-first content and its E-E-A-T guidance remain the most useful checklist available, and they are just as usable when no update has been confirmed at all. Google's core update guidance says plainly that there is nothing wrong to fix so. It still points you at those questions.

What to fix first, and what to leave alone for now

Do nothing destructive in the first fortnight. Deleting pages, bulk noindexing, mass redirecting: all of it removes content that frequently returns on its own, and all of it makes the next measurement impossible to read.

Work in this order:

1. Technical regressions. Indexing, canonicals, schema conflicts, anything you shipped. Fix immediately, these are not judgement calls.

2. Cannibalisation. Consolidate competing pages into one stronger URL and redirect the loser.

3. Depth on the pages that lost the most non-brand clicks. Not the pages that lost the most clicks overall. Non-brand.

4. Internal links. Last, once the target pages are actually worth linking to.

Rewrite for substance, not length. Adding 600 words of connective tissue to a vague page produces a longer vague page. Replace the generic paragraphs with figures you can source, process detail, named standards, and examples only your business could give. If you are also auditing partner or guest sections of the site, our walkthrough on auditing your own site for reputation abuse covers the sections most owners forget they host.

Resist the panic burst. Thirty refreshed or new pages live in a week produces an unnatural publishing spike and destroys your ability to attribute any recovery to anything. Judge nothing before four weeks. Review properly at eight. Treat the next confirmed core update rollout as the real test of whether your changes landed, because that is the point at which Google reassesses broadly.

Keeping your content programme steady through volatility

A repeatable weekly loop beats reactive scrambling every time: refresh keyword and competitor data, plan each brief against the live top ten rather than last quarter's assumptions, check existing pages so the new piece does not compete with one you already own, then publish on a ramped cadence instead of in bursts.

Run two quality gates, not one. An SEO check for structure, coverage and intent match, and a separate content integrity check for specificity, accuracy and voice. Where they disagree, integrity wins: a page that satisfies a checklist and says nothing is the exact profile that volatility punishes. That is the thinking behind how we write and grade every article before it reaches a client.

Keep a dated change log. Deployments, publishes, refreshes, redirects, plugin updates. The next unconfirmed episode is much easier to diagnose when you can compare it against a record rather than reconstruct it from memory and Slack scrollback. Awareness Marketing runs research, briefing, writing, publishing and reporting in one place for exactly this reason: a volatility week should cost you an afternoon of diagnosis, not a stalled publishing calendar.

Frequently Asked Questions

What is an unconfirmed Google update?

An unconfirmed Google update is a period of visible ranking movement that Google has not announced or documented. Google confirms only some changes, mainly core and spam updates listed on its Search Status Dashboard, and ships many smaller adjustments without comment. Third-party volatility trackers detect the movement, but nobody outside Google can state the mechanism.

How can I tell if the September 2026 volatility affected my site?

Compare matched, complete weeks either side of the suspected start date in Search Console, allowing for the two to three day data lag, and split brand from non-brand queries before reading anything into the totals. Then segment by folder, device and country. If non-brand clicks and positions fell across a whole section while brand queries held steady, you have a real organic movement worth investigating.

Should I change my content while an unconfirmed update is still rolling out?

Fix technical regressions immediately, and leave content edits alone for the first fortnight. Deleting or noindexing pages during unattributed volatility removes content that often recovers on its own, and it makes the next measurement unreadable. Use the waiting period to diagnose rather than to react.

How long does ranking volatility from an unconfirmed update usually last?

There is no published window, because the update is unconfirmed by definition. A sensible working rule is to judge nothing before four weeks, review properly at eight, and treat the next confirmed core update rollout as the genuine test of whether your changes worked.

Could AI-written pages be the reason my rankings dropped?

Google's guidance judges content by quality and usefulness rather than by how it was produced, so the question is not whether AI was involved. It is whether the pages are generic. Thin, templated writing with vague filler and nothing first-hand in it is the profile that tends to fall during volatility, whoever or whatever wrote it.