Google spam update and GSC image data: an AEO playbook

26 September 2026 · The Cresia team

Neither change asks you to rebuild your AEO or GEO programme. The September spam update is a reason to audit thin, scaled and borrowed content before it costs you citations. The new image data in Search Console is a reason to check that your visuals carry information and that your reporting can see them. Do both, in that order.

Key takeaways

Neither the spam update nor the new image data needs a new AEO strategy. Both reward pages that are useful, crawlable and clearly written.
The spam update may take up to two weeks, according to Search Engine Journal. Judge results against the date Google marks it complete, not the first day you see movement.
Audit pages by how they were made. Scaled, templated and hosted-for-others content is where the exposure is.
Images now have their own reporting in Search Console, so give them descriptive text, nearby context and crawlable URLs.
Measure by page group, keep a fixed prompt set for AI answers, and treat any single citation score as a trend line, not a verdict.

What Google's September spam update and the Search Console image data actually are

Search Engine Journal's SEO Pulse column, written by Matt Southern, reported three things together. Google's September spam update may take up to two weeks to roll out. Search Console has gained image-based search data. And there are new reports looking at traffic from Search and from AI Mode.

They're different kinds of news. The spam update is enforcement: Google adjusting how it catches content that breaks its spam policies. The Search Console addition is instrumentation, a new view of how people reach you when they start a search with a picture. The traffic reports are interpretation, other people's attempts to work out what AI Mode sends back to publishers.

A rollout that can run for two weeks won't hit every site on the same day. Rankings can swing early and settle later, and plenty of sites will see nothing at all. Google's Search Status Dashboard is where it normally marks an update as complete. That's the date to compare against, not the day you first noticed movement.

Why a spam update matters for AEO and GEO

Teams often run AEO as a separate workstream from SEO, with its own tools and its own slide in the monthly deck. The split is handy for reporting and misleading for cause and effect. Google's AI answers sit on top of Google's index and its quality systems. A page that's demoted or dropped for spam has one less route into those answers. Google hasn't published a formula linking spam actions to AI citations, so treat the link as likely rather than proven. It's still the safer assumption.

Picture a software company with a solid product-comparison section and, beside it, 600 near-identical location and use-case pages generated from a template with the nouns swapped. The comparison section is good. The template pages are the sort of scaled content Google's spam policies describe. If the update trims the template pages, nobody misses them. If it makes the whole domain look less trustworthy, the comparison section pays too. Which of those happens depends on how the site is judged, and you can't see that from outside. So you reduce your exposure.

GEO tactics carry their own risk. Mass-produced question-and-answer pages, text hidden from people but visible to crawlers, and manufactured reviews are all aimed at machines, and all are the kind of thing spam policies exist to catch. The tactic feels new. The violation is old. Read Google's spam policies on the Search Central site rather than trusting a summary, this one included.

How AI answer engines pick and cite sources

Engines differ, and none publishes its full selection logic. The general shape is consistent enough to plan around. The system takes a question, often splits it into several narrower searches (Google has described this fan-out approach for AI Mode), retrieves candidate pages, picks the passages that answer each part, writes a synthesis and attaches links to a handful of sources.

Three things follow from that shape.

First, retrieval comes before citation. A page that can't be crawled, rendered or indexed can't be cited, however good the prose. Second, passages get chosen, not pages. A paragraph that states its answer in the first sentence and stays on one question is easier to lift than one that builds slowly to a point. Third, the engine has to trust a source enough to name it. Clear authorship, facts that match across your own pages, and agreement with other reputable sources all help. Contradictions and unattributed claims hurt.

Nobody outside these companies knows the weights. A vendor selling a citation score to three decimal places is modelling, not reporting. Use those scores as a trend line, never as a verdict.

What to change during the rollout window

You have roughly two weeks in which conclusions are unsafe but preparation is cheap. Use them like this.

Pause bulk publishing. Two weeks without new templated pages costs almost nothing. Publishing 200 into a moving update makes your results unreadable.
Group pages by how they were made. Written by a named expert, edited from a source document, generated from a template, syndicated from elsewhere. You'll spot the risky groups in an afternoon.
Read a sample of each risky group. Take thirty pages and read them as a stranger would. If a page could be swapped for another with the nouns changed, it's a candidate to merge or remove.
Consolidate before you delete. Merge near-duplicates into one stronger page and redirect the rest. Remove only pages with no traffic, no links and no purpose.
Look at what you host for others. Partner sections, coupon subdomains, sponsored content and old acquired domains all borrow your reputation. Google's spam policies name abuses in this area, so check the current wording.
Rewrite your top pages for extraction. Put the direct answer in the first sentence of each section, name the entity, date the claim and show the author.
Keep a change log. Record every edit with a date. You'll need it when the numbers move and someone asks why.

The log matters more than it sounds. If you rewrite forty pages mid-rollout and traffic shifts, you won't know whether the update or your edits did it. Changing a small, labelled group and leaving a comparison group alone gives you something you can read.

Using the new image search data in Search Console

Image-based search means people starting with a picture rather than words. Exactly which queries and surfaces the new data includes, and how it's broken down, is defined in Google's Search Console documentation. Read the help page before you build a report on it.

Direction matters more than definition at first. If image-led discovery is a growing way in, the visuals on your pages are part of how you get found, not decoration. That has practical consequences:

Make images say something. A product photo that shows scale, a labelled diagram or a real screenshot carries more information than a stock photo of a handshake.
Put words near them. Descriptive alt text, a caption, and a nearby sentence that names what's pictured. A filename like img_4471.jpg helps nobody.
Keep them crawlable. Image URLs shouldn't be blocked in robots.txt, and images that only load after a script interaction may never be seen.
Match structured data to the page. Where product markup includes image properties, keep it consistent with what a visitor actually sees.
Know your asset library. The same image in five sizes across three domains with no naming convention makes it hard to tell which version is being found. Creative teams handling this at volume will recognise the problem, and the Mosaic page covers Cresia's take on creative management.

Start with one page type. Pick your top product category, read its image data, and fix its images before you generalise. Expect the numbers to be small for most sites at first. Small isn't ignorable, but it doesn't justify a whole programme either.

Measuring AEO and GEO after these changes

Measurement is where AI search work most often goes soft. Some assistants strip referrers, some answers give no click at all, and answers vary from one run to the next. So build a plan that admits those limits instead of hiding them.

Question	Where to look	Owner	When
Did the spam update change our organic traffic?	Search Console performance, split by page group	SEO lead	Two weeks after Google marks the update complete
Are images bringing visits?	The new image-based search data in Search Console	SEO lead with creative	Monthly
Are we cited in AI answers?	A fixed prompt set, checked by hand or with a visibility tool	AEO/GEO lead	Monthly
Do AI referrals show up as their own channel?	Analytics channel rules and referrer list	Analytics lead	Quarterly
Did any of it affect leads or revenue?	Conversions from landing pages in the affected groups	Growth lead	Monthly

Compare like with like: the two weeks before the update began against the two weeks after it finished, same weekdays, same page groups. Sitewide totals hide the story, because a good section can mask a bad one.

For AI answers, fix a list of 30 to 50 prompts that reflect how buyers actually ask, and record the date, the engine and whether you were cited. Repeat monthly. It's tedious and it works. If you'd rather not do that by hand, the MediaPilot page covers Cresia's media and AI-search visibility product.

And check the plumbing. If your referrer classification lives in a spreadsheet nobody owns, AI-driven visits will be filed under direct or referral and your trend lines will lie. A tracking specification is the plain fix: one written source of truth for how channels are defined.

For how AI Mode clicks are counted inside Search Console totals, read Google's own documentation rather than assuming.

What to ignore and what not to do

Most of the noise around a spam update is anecdote. A few habits will save you from acting on it.

Don't rewrite everything because of two weeks of volatility. Uneven rollouts produce dramatic screenshots and little information.
Don't answer a spam update with more content. If the problem is scaled, thin pages, adding more of them is the wrong direction.
Don't write for crawlers in ways people can't see. Hidden text and prompt-style instructions aimed at models belong on the wrong side of the policies.
Don't treat one tool's citation score as fact. Different tools sample prompts differently, so compare a tool to itself over time.
Don't chase week-one winners and losers on social media. Wait for the completion date and your own data.
Don't skip the boring checks. A spike in image 404s or a robots.txt change can look like an algorithm problem and be nothing of the kind.

The position that holds up: fewer, better pages, each answering one question plainly, with images that show something and analytics that can see where visitors came from. Everything else is packaging.

Frequently asked questions

How long will the September spam update take?

Search Engine Journal reported that Google said it may take up to two weeks. Check Google's Search Status Dashboard for the completion notice and compare your data from that date, not from the first day you noticed a change.

Will the spam update affect my visibility in AI answers?

Possibly, indirectly. Google's AI features draw on the same index and quality systems as classic search, so a page demoted for spam has fewer ways into those answers. Google hasn't published how the two connect, so watch your own page-group data instead of assuming.

Do I need to optimise images separately for AI search?

Not as a separate programme. Descriptive alt text, captions, nearby context and crawlable URLs help images in every setting. Use the new Search Console image data to see whether it's worth going deeper on a given page type.

What should I do if traffic drops during the rollout?

Wait until Google marks the update complete, then compare by page group against the weeks before it began. If a group of scaled or thin pages fell, review it against Google's spam policies and consolidate or improve it. Avoid reverting edits blindly, and keep in mind that recovery, where it's possible, isn't always quick.

Sources

https://www.searchenginejournal.com/seo-pulse-google-spam-update-image-search-data/590882/
https://developers.google.com/search/blog
https://searchengineland.com/topic/generative-engine-optimization