AI Mode in Google Discover is not a toggle you can switch off, and it is almost certainly not the reason your blog traffic moved last month. It matters anyway. Discover and AI Mode used to be separate doorways: one a feed of cards in the Google app, the other a conversational Search surface. They now share entry points, which means a tap that once opened your article can open a generated answer instead. For a small business blog that gets a meaningful slice of its sessions from the feed, that changes what a click is worth and how you should read your own numbers.

Most coverage of this treats AI Mode as a feature review: where the toggle lives, which Labs setting enables it, what the interface looks like. Useful if you are a user. Useless if you are the publisher on the other end of the tap.

What AI Mode inside Google Discover actually is

Discover is the personalised feed you see in the Google app and on the new tab page in Chrome for Android. It is entity-driven, not query-driven: Google decides what to show you based on interests, history and location, and nobody typed anything. That single fact breaks the usual SEO workflow, because there is no keyword to rank for, no average position that means much, and no keyword gap to close.

AI Mode is the other thing. It sits inside Search, it generates an answer, and it invites follow-up questions in the same thread. Where AI Overviews appear above the classic blue links and leave your result intact underneath, AI Mode is the destination. The tap ends there unless the reader chooses to go further.

The change worth noting: Search Engine Roundtable reported on 23 December 2025 that some Google app notifications, which look like news alerts, open an AI Mode response rather than a publisher page. Treat that as what it is, an observed rollout at a point in time, not a permanent architecture. Google tests entry points constantly and removes them just as quietly. Check the current behaviour in your own app before you build a theory on it.

The publisher-side distinction is simple. AI Overviews cost you some clicks at the top of a results page. An AI answer sitting where your article used to open costs you the visit entirely, and it costs you the second and third page view that followed it, because the follow-up question now happens inside Google rather than on your site.

How this changes the traffic a small business blog can expect

Discover traffic was never stable, and anyone who has watched it for a year knows the shape. A post sits flat for six weeks. Then the feed picks it up, and a single URL can account for the large majority of a month's sessions inside a 48-hour window, overwhelmingly on mobile, because Discover is an in-app surface. Then it stops. A UK trades or services blog with one lucky post sees exactly this, and the founder reads the collapse as a penalty. Nothing was penalised. The feed simply moved on.

Where AI Mode absorbs the first tap, the damage shows up in a specific order. Sessions per reader fall first. Clicks fall second. If you only watch clicks, you will miss a month of drift.

Some content is far more exposed than the rest:

• Most exposed: news-reactive posts, definition pieces, "what is X"explainers, and listicles that a generated response can summarise in full without losing anything. If a model can reproduce your article from training data and a crawl, it will.

• Least exposed: first-hand specifics. Real UK prices in £, dated observations, named tools with version numbers, your actual process, a screenshot of your own dashboard, the mistake you made in March and what it cost to fix.

Here is the part that connects two problems people treat separately. The content most likely to be summarised away is the content that already reads as machine-written: flat generic definitions, even sentence rhythm, no first-hand detail, nobody in particular behind the words. Countering AI Mode and countering AI tells is the same editorial job. We go into the mechanics of that in how we research and write every brief, but the short version is that specificity is the defence, and there is no other one.

Reading your own numbers: Discover, AI Mode and Search Console

Most SMEs blame AI Mode for a Discover drop without ever opening the right report. The two surfaces fail for completely different reasons, and Search Console keeps them apart.

Discover has its own report, separate from the Search performance report. It only appears if your site has met the minimum threshold for feed traffic in the past 16 months. It gives you clicks, impressions and CTR, filterable by page, country and device. It gives you no query data at all, because there were no queries. Impressions are counted when your card is viewed in the feed, not when it is tapped.

AI Mode is not in that report. Google folds AI Mode clicks and impressions into the Web search type inside the Search performance report. There is no clean AI Mode line to isolate and subtract. If you want the detail on what that means for reading the Search side, we covered it in how to read AI Mode data in Search Console.

A 20-minute diagnostic that actually tells you something:

1. Open the Discover report, set the range to the last three months, and note clicks, impressions and CTR. Export the top five URLs.

2. Switch to the Search performance report, Web search type, and filter to those same five URLs. Compare clicks and average position over the same window.

3. Open GA4, look at landing page sessions for those URLs, then check the device split and sessions per user. Discover referrals typically arrive through the Google app rather than a browser, so the device and source breakdown is your sanity check.

The read is straightforward. Discover clicks down, Web clicks and average position flat: the feed stopped showing you, and your content quality is not the variable. Both down together: that is a search-side problem, and it probably predates AI Mode. Discover flat, Web clicks down while impressions hold: that is the pattern consistent with answers being satisfied before the click.

What to change in the publishing plan, and what to leave alone

Do not rebuild your blog around Discover. We say this to clients who have just had their first feed spike and want twelve more posts like it: the feed is unforecastable, there is no keyword to target, and a business that reorganises its editorial calendar around it has traded a plannable channel for a lottery ticket.

Plan against live search demand instead. We brief every piece against the top 10 results that are actually ranking for the target term, which is how you know what the page has to cover to compete, and then treat feed pickup as upside. The full workflow behind that planning process is published if you want the detail.

Three things to add to posts that are exposed:

• Numbers only you have. A quoted price, a job duration, a conversion figure from your own site, dated so the reader knows when it was true.

• Named specifics. The tool, the version, the setting, the plugin, the standard. "A scheduling tool"is summarisable. "Simpro's recurring job scheduler, which we moved off in 2024 because of X"is not.

• Rhythm. Vary sentence length deliberately. Uniform cadence is the tell that makes writing read as machine-generated, and it is also what makes a post trivially easy to compress into three bullet points.

Two things not to do. Do not publish more in response to a dip: a site that published monthly and suddenly fires out fifteen posts in a fortnight looks like a spam pattern, not momentum, and the ramp should be gradual. And do not write a second post on an angle you have already covered. Two pages competing for the same intent split authority between them, which is the worst possible move at the moment clicks are scarcer. Check for duplicate titles and overlapping targets before you commission anything, not after.

Technical housekeeping that still earns the surface

Google's Discover documentation in Search Central is short and worth reading in full. Two points matter most for small sites. Large images, at least 1,200px wide, enabled for large preview, are what get a post the visual card treatment. A 900px hero image is a common and entirely self-inflicted reason a post never gets picked up. And clickbait titles are named as a problem: a bait headline pulls a tap and then an immediate bounce, which is precisely the behaviour you do not want associated with your domain.

On structured data: detect what you already have before you add anything. If Yoast or a similar SEO plugin is already emitting Article markup and your theme or publishing tool adds a second block, you end up with conflicting entities on the page. One clean Article and Breadcrumb set beats two competing ones. Run the URL through Google's Rich Results Test first.

Core Web Vitals carry more weight here than in classic search for one practical reason: Discover traffic is almost entirely in-app, on mobile, on a connection you do not control. A layout that shifts while a reader is deciding whether to stay loses them. Google's own Web Vitals guidance covers the thresholds.

A 30-day plan for watching Discover without panicking

Week one. Baseline it. Discover clicks, impressions and CTR for the last three months, plus the five URLs responsible for most of it. Write the numbers down somewhere outside Search Console so you are comparing against a record, not a memory.

Week two. Tag those five URLs: summarisable or first-hand. Rewrite the weakest one with concrete specifics, real figures and dated detail. Do not rewrite all five.

Weeks three and four. Publish two intent-led pieces aimed at search demand, not at the feed. Note whether Discover picks either up, and do not draw conclusions from a sample of two.

Decide your trigger in advance, in writing. A sustained Discover drop with flat Web clicks is a feed decision and needs no response. A drop on both surfaces over six to eight weeks is a content and technical problem, and that is when you act. If you want an outside read on where your pages currently stand before you start, our free content marketing snapshot is the quickest way to get one.

AI Mode in Google Discover is a real shift in where the tap lands. It is not an emergency, and it is not a reason to abandon a publishing plan that was working. Write the things a generated answer cannot carry, measure the two surfaces separately, and let the feed be a bonus rather than a strategy.

Frequently Asked Questions

Is AI Mode in Google Discover the same thing as AI Overviews?

No. AI Overviews appear above the classic results on a search page and leave your listing in place below them. AI Mode is a conversational Search surface where the generated response is the destination, with follow-up questions handled in the same thread. The publisher impact differs: an Overview costs you some clicks, while an AI Mode entry point can replace the visit altogether.

Can I see AI Mode traffic separately in Google Search Console?

Not as its own line. Google folds AI Mode clicks and impressions into the Web search type within the Search performance report, so there is nothing to isolate or subtract. Discover, by contrast, has a completely separate report with no query data. If you want to know which surface moved, you have to compare the two reports for the same URLs rather than looking for an AI Mode filter.

Will Discover traffic disappear for small business blogs?

There is no evidence pointing that way, and we would be cautious about anyone who claims to know. What is reasonable to expect is that Discover becomes less reliable as a single entry point, because some taps now resolve inside Google. The sensible position is to keep Discover as upside and build your plan on search demand you can actually forecast.

Should I change my blog strategy because of AI Mode in Google Discover?

Change the content, not the strategy. Add the things a generated answer cannot reproduce: dated first-hand observations, real UK prices in £, named tools and settings, your own process. Keep planning against live search results for terms people actually type, and do not reorganise an editorial calendar around a feed nobody can predict.

Why does my Discover traffic spike and then vanish within days?

Because that is how the feed behaves. Discover surfaces a card to a slice of users based on interests rather than a query, and when that cycle ends the traffic ends with it, often inside 48 hours. A collapse after a spike is almost never a penalty. Check whether your Web clicks and average position stayed flat over the same period: if they did, the feed moved on and your content is fine.