Google Web Guide Classic Search Bug: What AEO Teams Do

21 September 2026 · The Cresia team

Not reliably. Search Engine Roundtable reports that in Google's Web Guide beta, the button labelled Classic search doesn't take you out of Web Guide and back to the ordinary results page. For AEO and GEO teams the bug itself matters little; what matters is that Google is running several result layouts at once, and your visibility is now split across them.

Key takeaways

Web Guide is an opt-in beta that blends AI-style and web-style results, and Search Engine Roundtable reports its Classic search button doesn't exit it.
One bug in one beta is not a traffic event. The pattern behind it, several layouts live at the same time, is the part to plan around.
Ranking, retrieval and citation are separate jobs. Pages that do all three hold up under any layout.
The work that pays off is dull: extractable answers, clear entities, crawl access and honest structured data.
Measure by surface and by query group. A single rank number no longer describes what a searcher saw.
Ignore screenshots of the newest interface until it stops being a beta.

What is the Web Guide Classic search bug?

Web Guide is an experimental way of using Google Search that people opt into. Search Engine Roundtable describes it as a mix of Google's AI results interface and the traditional web results interface in one place. Inside it sits a button labelled Classic search, which you would expect to drop you into the standard results page.

According to Search Engine Roundtable, it doesn't. Click it and you stay in Web Guide.

That's the whole story, and it is worth keeping the size of it in view. It's a beta. It's opt-in. A bug in an escape hatch tells you the feature is unfinished, not that it's about to reach everyone. Google has changed, paused and folded experiments into other products before, and nothing in the report suggests a rollout date.

What the report does confirm is something you may already have noticed in your own searching: there isn't one Google results page any more. There's the classic list, AI Overviews, an AI Mode, and experiments like this one. Search Engine Roundtable also covered Google testing an AI Mode button on the search bar results page, which points the same way.

Why does one beta bug matter for AEO and GEO?

Because it undermines an assumption most reporting is built on: that a keyword has one position.

Imagine a growth lead at a mid-size software company who tracks 300 non-brand queries every Monday. The rank tracker says the pricing-comparison page sits at position four. Meanwhile a colleague who opted into a beta sees that same query returned as grouped clusters of results with AI-written summaries above them. The page might be in one cluster, in none, or cited in a summary. The tracker reports the same thing either way.

That gap is the practical issue. It doesn't require Web Guide to reach everyone. It only requires that a meaningful group of your buyers sees something other than the layout your tools assume. Analysts, developers and early adopters tend to be the ones who opt into betas, and those are often exactly the people who research and recommend software inside a company.

My position: don't rebuild your strategy around Web Guide, and don't dismiss it either. Use it as a prompt to check whether your work depends on one layout staying put.

How do AI answer engines choose and cite sources?

The mechanics vary by product and change without notice, so treat any tidy formula with suspicion. What holds across most systems is a rough three-stage flow.

Access. The system has to be able to fetch your page, or have it in an index it can query. Blocked crawlers, login walls, heavy client-side rendering and slow servers all remove you here, before quality is even considered.
Retrieval. Given a question, the system pulls candidate passages, often more than one query variant per question. Search Engine Journal ran a piece on checking whether a page is part of a retrieval pipeline for AI, which is the right question to ask about your own URLs.
Selection and citation. The model writes an answer and attributes some of the material. Passages that state a claim plainly, name their subject and stand alone without surrounding context are easier to lift than ones that depend on three earlier paragraphs.

A page can fail at any stage independently. A page that ranks well can be poorly extractable. A page that is beautifully written can be blocked by a robots rule someone added two years ago. Search Engine Journal's coverage of Cloudflare generating robots.txt files is a useful reminder that crawler access is increasingly set by defaults and platforms, not by a person reading the file.

Web Guide-style layouts add a fourth consideration: grouping. When results are organised by theme, the question becomes which theme your page is filed under. A page that tries to cover six angles loosely may not land clearly in any of them.

What should you change on your pages now?

Nothing here is specific to Web Guide, and that's the point. These changes help under every layout Google is currently testing.

Put the answer first. Each section should open with a direct answer to the question its heading asks, then explain. Someone quoting one passage should not need the rest of the page.
Name things in full. Write the product, the company, the standard or the metric out in the passage instead of saying it, this or the tool. Passages get lifted out of context; pronouns don't survive.
Give each page one clear job. If a page could be filed under three topics, split it or pick a lead. Grouped result layouts reward pages that are obviously about something.
Match structured data to the visible page. Mark up what a reader can actually see: organisation, article, product, FAQ where you truly have questions and answers. Don't add markup for content that isn't there.
Audit crawler access. Read your robots rules, CDN bot settings and any firewall rules as they are today. Decide deliberately which AI crawlers you allow, and write down why.
Keep facts current and dated. Pricing, feature lists and comparison pages that quietly go stale are the easiest way to be cited wrongly.
Check that the page renders without JavaScript doing the heavy lifting. If the main content only appears after scripts run, some fetchers will see an empty shell.

If you own the paid and organic side together, the media teams solution page describes how that joint view works in practice, and MediaPilot is the place to look for AI-search visibility tracking.

How do you measure visibility across blended results?

Start by admitting what your current numbers can't see. Rank trackers report a position on a layout. Analytics reports sessions after a click. Neither reports being named in an answer where nobody clicked.

A workable plan keeps the old numbers and adds a few honest ones next to them.

Question	Signal	Where to look	Owner
Are we crawlable by the systems that matter?	Fetch success and blocked-bot rules	Server logs, CDN and robots settings	Technical SEO
Are we named in AI answers for our core questions?	Mention and citation presence by query group	Repeated manual checks or a visibility tool, logged with date and surface	SEO or AEO lead
Is the wording about us accurate?	Correct or incorrect description of product, price, category	Same checks, scored by a person	Content lead
Do those visits convert?	Sessions and conversions from AI-referred sources	Analytics, with referrers grouped consistently	Analytics lead
Did a layout change move us?	Before/after on a fixed query set	Dated snapshots	SEO lead

Three habits make this useful. First, fix a query set of 30 to 50 questions that map to real buying decisions, and don't change it every month. Second, record which surface you checked, because the same query can look different in classic results, an AI Overview and a beta layout. Third, keep the referral naming consistent. If different people label the same source three ways, your trend line is fiction. A shared tracking specification is the simplest fix.

Expect noise. AI answers vary between runs, so one check proves little. Look for direction over several weeks, and be wary of any tool that reports precision it can't have.

What should you ignore, and what should you avoid?

There's a temptation, every time a new interface appears, to treat it as a reason to rewrite everything. Resist it.

Ignore:

Screenshots and hot takes about which layout will win. Google is testing several and hasn't said.
Claims that a specific trick gets you into a beta interface. Nobody outside Google knows how these groupings are built.
Advice to write differently for each surface. You'll produce five thin variants of one page.

Avoid:

Blocking crawlers as a reflex. Some publishers have good commercial reasons to restrict AI access. Make it a decision with an owner, not a leftover rule.
Markup for its own sake. Schema that describes content the page doesn't contain can hurt trust and helps nothing.
Stuffing pages with question headings. Real questions with real answers are good. Forty rephrasings of one question read as padding.
Reporting one blended score to leadership. It hides which surface moved. Show the table above instead.
Waiting for a stable interface. There won't be one. The teams that cope are the ones whose content is clear enough to survive a redesign.

What does this say about where Google Search is heading?

The direction is fairly clear even if the details aren't. Google is placing AI-generated organisation on top of, or next to, ordinary results and letting users move between the two. Web Guide, AI Mode buttons on the search bar and AI Overviews are different attempts at that same blend.

For a team, that has an organisational consequence more than a technical one. SEO, content, analytics and paid can no longer each own a separate slice of the results page. Somebody has to hold the shared view: which questions matter, which surfaces we check, how we describe results consistently. Under the umbrella of marketing operations, that's a coordination problem, and coordination problems get solved with a named owner and a fixed cadence.

A reasonable cadence: a monthly review of the fixed query set, a quarterly crawler and markup audit, and a standing rule that anyone who sees an odd layout logs a dated screenshot to a shared folder. It costs an hour or two a month and gives you an evidence trail when leadership asks why a number moved.

The bug itself will probably be fixed or forgotten within weeks. The habit of checking more than one surface is the thing worth keeping.

Frequently asked questions

Should we optimise specifically for Google Web Guide?

No. It's a beta that people opt into, and Google hasn't committed to a rollout. Work on answer-first pages, clear entities and crawl access instead, since those help in Web Guide, AI Overviews, AI Mode and classic results alike.

Does the Classic search button bug affect our rankings?

There's no indication that it does. Search Engine Roundtable reported a navigation bug inside the interface, not a ranking change. It's relevant only because it shows that people can be stuck in a non-classic layout, which your rank tracker doesn't reflect.

How do we know if an AI answer engine is citing our pages?

Run a fixed set of real buyer questions on a schedule, record which surface you used, and log whether your brand or URL is named and whether the description is accurate. Repeat over several weeks, since answers vary between runs. Server logs and referral data add a second angle on access and visits.

Is structured data enough to get cited?

No. Markup helps systems understand what a page is, but access, retrieval and the quality of the passage decide whether it gets used. Treat schema as one input, and keep it consistent with what's visible on the page.

Who should own AI search visibility inside a marketing team?

One named person, usually the SEO or growth lead, with input from content, analytics and paid. The job is to hold the query set, the checking cadence and the reporting format. Without that owner, each function measures its own slice and nobody sees the whole result.

Sources

https://www.seroundtable.com/google-web-guide-bug-42124.html
https://www.seroundtable.com/google-search-bar-testing-ai-mode-button-42110.html
https://www.searchenginejournal.com/checking-a-page-is-part-of-a-retrieval-pipeline-for-ai/589284/
https://www.searchenginejournal.com/cloudflare-will-write-your-robots-txt-and-it-has-a-point/589262/