Quick Answer
AI Overviews are Google's AI-generated summaries, shown above the standard results and built from pages retrieved out of the same Search index. Google reports over 2.5 billion monthly active users for them. The founder response is narrower than the panic suggests: confirm Search Console still includes your site, keep your pages indexed and snippet-eligible, publish the things only you can write, then track impressions in the Generative AI performance report.
Most of what you have read about AI answers in Search was written for publishers. Publishers sell attention at scale, so a fall in outbound clicks takes away the thing their whole model runs on.
You are not running a media business. Your site has narrow demand, a small audience and buyers who already half know what they need. That changes the arithmetic, and it changes what is worth doing in the next ninety days.
Something changed on Google's side in 2026 too, and it gives you an instrument rather than a theory. This covers the four checks to run in Google Search Console, the content changes Google's own guidance asks for, the list of things it says you can stop doing, and what to measure once clicks stop telling the whole story.
Google Now Answers Above the Links It Shows
Google now places a generated answer above the standard results on a large share of queries, and it builds that answer from pages in the same index that produces those results. AI Overviews passed 2.5 billion monthly active users and AI Mode passed one billion, on Google's own count, published in June 2026 and updated on 31 August.
The retrieval has a name. Google calls it retrieval-augmented generation: its ranking systems fetch relevant pages, and the model writes the response from what those pages say.
A second mechanic matters more for a small site. Query fan-out turns one question into several concurrent queries, so your page can be pulled into an answer for a question the user never typed.
Scale arrived faster than the advice did. Elizabeth Reid, Google's VP of Search, put it this way at I/O on 19 May 2026. "Just one year after its debut, AI Mode has surpassed one billion monthly users, with queries more than doubling every quarter since launch."
Google also says it has added more inline links inside responses, plus website previews, to push people through to the page. That is Google describing its own product. No independent measurement confirms it.
How much traffic actually disappears is a separate argument with its own numbers. The honest version of it is in what the 68 percent zero-click figure really measures, and this post does not relitigate it.
Your Four-Step Response Plan for AI Overviews
Run four steps in this order: confirm Google still includes your site, confirm the page can be shown with a snippet, change what you publish, then measure it. The order matters, because the first two are switches and no amount of writing survives either one being off.
Each step takes under an hour, and none of them needs a developer unless step two turns one up.
1. Confirm Google Still Includes Your Site
Open Google Search Console, go to Settings, then Search generative AI. The control has three values: include your site's links and content, exclude them, or inherit the setting from a parent property.
Include is the default for every property. Google announced the control in June 2026, and Search Console's own documentation records it reaching all websites worldwide on 31 August 2026.
Excluding is not a soft signal. Google states that a site which opts out gets no traffic and no impressions from those features, and that the control is not used as a ranking signal anywhere else in Search.
The mistake is assuming the default holds. A URL-prefix property inherits from its domain property. So whoever switches the parent switches everything beneath it, quietly and without a warning email.
You know it worked when the control page reads include, on the exact property you publish from. Changes take one to two days to take effect.
2. Confirm the Page Can Be Shown With a Snippet
Google's eligibility rule runs to one sentence. To appear in generative AI features, a page must be indexed and eligible to be shown in Google Search with a snippet. That makes three robots directives load-bearing in a way most teams never audit.
A nosnippet meta tag, a max-snippet:0 directive or a stray data-nosnippet attribute wrapped around your main content removes eligibility without removing the page from Search. Check the served HTML, not the component that generates it.
On a JavaScript site the served HTML is a different artifact from what you see in your browser, which is the trap laid out in Googlebot renders React and AI crawlers do not. Run URL Inspection on your three most important pages and confirm each one is indexed.
If one is not indexed, nothing downstream of this step is worth doing yet. An unindexed page cannot appear in AI Overviews, in AI Mode, or in the ten blue links either.
3. Publish the Non-Commodity Half of What You Know
Google draws this line with an example rather than a rule. Commodity content is the kind it describes as "7 Tips for First-Time Homebuyers", common knowledge that could have come from anyone and adds little for a reader.
Its counter-example is "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line", which nobody else could have written. For a software founder the equivalent sits in your own logs and invoices.
The migration you ran and what broke in week three. The benchmark you measured on your own hardware, with the configuration attached. The pricing you actually charge, and why it changed last year.
That is also where Better Marketing starts an AI SEO service engagement, because a page nobody else could have written is the only durable input an answer engine has to work with.
The common mistake is the opposite reflex: spinning up a separate page for every fan-out variation of the question. Google names that as scaled content abuse and says outright that a high quantity of pages does not make a website higher quality or more relevant.
You know it worked when the page starts appearing in the Pages view of your Generative AI performance report, and when someone forwards it to you unprompted.
4. Measure Impressions in the Generative AI Report
Search Console's Generative AI performance report launched on 3 June 2026 and reached every website worldwide on 31 August 2026. It reports impressions from AI Overviews and AI Mode, broken out by page, country, device and date.
It does not report clicks, click-through rate or queries. Google Discover has its own separate generative AI report, so do not go looking for Discover numbers inside the Search one.
Take a reading before you change anything. Without a baseline the first month of data is a number with nothing to compare it against, which is how a quarter of work ends up unprovable.
What Google Says You Can Stop Doing Now
Google published a list of things that do not help in generative AI search, and several of them are exactly what the market has been selling since 2025. Reading it is the cheapest hour in this plan.
- llms.txt and other machine-readable files. Google Search ignores them, and says creating one will neither harm nor help your visibility or rankings.
- Chunking content into tiny pieces. There is no requirement to break a page up, and no ideal page length.
- Rewriting pages in a special voice for AI systems. They understand synonyms and general meaning, so the long-tail variants are unnecessary.
- Seeking inauthentic mentions across the web. The same systems that block spam in core results feed these features.
- Overfocusing on structured data. It is not required for generative AI search and there is no special schema.org markup to add, though it still earns rich results elsewhere.
None of that makes structure worthless. Google still asks for paragraphs, sections and headings that give a reader a clear way to navigate a page, and none of it changes your eligibility for AI Overviews, which stays a question of indexing and the snippet rule.
A lot of content marketing for tech companies has been rebuilt around those five tactics over the past year, which is an expensive way to produce nothing. Google names the industry behavior itself, telling site owners weighing third-party advice or services to check it against official guidance first. The full list sits in Google's guide to optimizing for generative AI features, last updated on 10 July 2026.
How AI Impressions Differ From the Clicks You Tracked
AI impressions differ from the clicks you tracked in what they count and in what they refuse to tell you. The dedicated report is a visibility instrument, not a performance one.
| Factor | Generative AI performance report | Standard performance report |
|---|---|---|
| What it counts | Impressions in AI Overviews and AI Mode | Impressions, clicks, CTR and average position |
| Clicks and click-through rate | Not shown | Shown |
| Query data | Not available | Available |
| Dimensions offered | Pages, countries, devices, dates | Pages, countries, devices, dates, queries |
| Google Discover | A separate generative AI report | The standard Discover report |
Read the Pages tab first and sort by impressions. The pages at the top are the ones Google is already willing to put in front of an answer, which makes them the cheapest things in your library to improve.
Then hold the limit in view. An impression is not demand, a page can appear in a hundred answers and produce nothing the report can see, and nobody can hand you attribution when the data carries no click and no query.
If you are hiring for this, it is a fair question to put to any B2B demand generation agency: name the pages you will move, and say what the Generative AI performance report should look like in ninety days. An agency that answers with a citation count is quoting a number Google does not publish.
When Chasing AI Overviews Is the Wrong Priority
Chasing AI Overviews is the wrong priority when the problem sits upstream of search altogether. Three cases come up repeatedly, and all three are about sequencing rather than about the tactic itself.
The first is a product nobody is searching for yet. If the category has no query volume there is no answer to be included in, and the work belongs in founder-led writing and direct conversation instead.
The second is a site that is not indexed at all. Crawling and indexing come first, and nothing in this plan applies until that is true.
The third is timing. A company whose pipeline still runs on two warm introductions a month gets more from the next twenty conversations than from a quarter of publishing. In that case a B2B demand generation agency that opens with outreach rather than articles is the right call.
There is a limit in the method too, worth saying before somebody sells you past it. We cannot prove a content change caused an inclusion in AI Overviews or in AI Mode, because the report gives impressions by page and no attribution at all.
Once the sequencing is right, the next question is how to become the source rather than merely stay eligible, which is the ground AEO versus SEO for startups covers. A serious AI SEO service will tell you which of those three cases you are in before it sells you anything.
Run the Four Checks Before You Publish Again
The change in Search is real, and it is not the same change for everyone. Publishers lost a traffic model you were never running in the first place.
Open Search Console, confirm the Search generative AI control still reads include, confirm your strongest page is indexed and snippet-eligible for AI Overviews, then publish the one thing in your company nobody else could have written.
Measure it ninety days later. That sequence is most of what content marketing for tech companies now amounts to, and it is the work Better Marketing does with founders taking a product to market.