Google AI Overviews: How to Check the Impact on Traffic and Leads

Your article still ranks, but fewer people visit it. Google now shows an AI answer above some results. Marketing sees fewer clicks; sales does not yet know whether qualified enquiries have changed. A budget decision needs both sets of evidence.

In brief: AI Overviews do not impose a fixed traffic loss on every website. Compare queries, devices, impressions, clicks and leads. Then check the content and access of specific pages. A citation, a link impression and a sale are separate measurements.

Canem Errant checks the questions buyers use, the observed search results and the documents linked in the answer. We choose a limited change and repeat the observation. That supports a decision about content or a paid test without promising a percentage of traffic back.

What are AI Overviews, and how can they change the buying journey?

AI Overviews are Google’s generated summaries with supporting links. SGE was the earlier experimental name. An overview can satisfy a question or direct someone to more detail; it does not appear for every query. Google describes query fan-out, which explores related questions and sources. That does not reveal the exact evaluation of every document. Google AI features.

A person reading a definition may finish their search. Another needs implementation conditions, compatibility, prices or evidence about a supplier. Your material should help that decision. A visit after an overview does not automatically produce a better lead: qualification and deals must be measured.

Five query types to include in your check

Keep different buying intentions in the sample. This table defines what to observe, not how often an overview will appear.

Record language, country, device and date for each query. A restaurant question does not represent a B2B purchase; an English check does not replace German demand research. Preserve the market and task when comparing results.

Why does a technical GEO score not prove visibility?

An internal Canem Errant note dated 13 July 2026 reported checking 14,718 websites, requesting the homepage, robots.txt and llms.txt. Its historical labels included 6,474 WARM-GEO and 685 HOT-GEO sites. The original per-site records and the treatment of scan errors are not presented here. These are reported internal classifications, not an independently verified study of Google recommendations.

The figures do not show that 685 companies lost AI traffic or that removing a restriction immediately increases sales. A file inventory is not Google’s ranking formula. The GEO guide gives the wider context; a useful diagnosis still needs a specific barrier and a repeatable question.

Six areas to check instead of a promised 100-point ranking

1. Search access: the actual URL, server response, robots.txt and bot protection.
2. Display controls: applicable meta tags, HTTP headers and the Search Console AI inclusion setting.
3. Additional formats: whether a specific consumer needs llms.txt; Google Search does not require it.
4. Structured data: whether organization, address, author and service match visible facts.
5. Questions and answers: whether buyers can find clear conditions for a decision.
6. Content: useful first-hand information and traceable sources rather than general promises.

Keep search access separate from model training. Google-Extended does not determine inclusion in Google Search; GPTBot concerns possible OpenAI training. OpenAI’s OAI-SearchBot serves its search features. They are not one “AI blocker.” A noai tag is not a documented Google AI Overview exclusion switch. Google crawler documentation, OpenAI crawler documentation.

Four ways to improve an important page

1. Make the heading promise a useful answer

“How do I verify payment and order delivery to Meta?” tells a reader more than “Modern technology.” Use questions when they help. There is no requirement to turn every H2 into a question or publish every possible wording of a query.

2. Put the answer, conditions and evidence together

Answer directly, then explain applicability. A buyer needs to know what the diagnosis includes, which data are required and what will be changed. A case needs its market, period and scope. If compatibility depends on a CMS plan, state that. Clear limitations help the reader decide.

3. Make markup describe the content

Check appropriate article, author, organization and business types. Do not add unseen certifications, reviews or prices to strengthen machine-readable claims. Google requires neither a special AI schema nor llms.txt; your technical brief should start with an actual problem. Current Google guidance.

4. Make authorship and original information verifiable

Explain who is responsible and what supports the conclusion. Use authorized case data, documented tests and primary sources. E-E-A-T is useful for discussing experience and trust; it is not a public formula for buying a citation. Original figures without a method are less useful than a clear explanation with evidence.

What if informational pages lose clicks?

Check whether the loss occurs on queries where you actually observe overviews. Account for seasonality, query mix, snippets and device-specific positions. Three content directions are worth considering:

• Specific questions: implementation, compatibility, limitations and verification. Complexity can make a question valuable but does not guarantee a visit.
• Choosing a solution: comparisons, service scope, cost and a clear way to discuss the problem.
• Original evidence: approved cases, studies with methods and worked examples that support a buying decision.

Do not delete useful definitions merely because clicks fall. They can help buyers understand a service and navigate to detailed material.

Illustrative scenario: a B2B blog. If introductory visits fall, compare enquiries across the content group and test one page about product compatibility. Keep the original data. This is a test design, not a client case promising a 25% conversion increase.

Illustrative scenario: a professional practice. If an overview summarizes a consultation topic, check the author, source accuracy and contact conditions. A citation may support trust, but actual enquiries establish the business result. Years of experience and publication counts require that practitioner’s records.

How should you balance organic content and advertising?

Fewer informational clicks alone do not justify moving budget to Google Ads. Check qualified enquiries, affordable acquisition cost and sales capacity. A campaign is a separate hypothesis with a spending limit. Google Ads vs SEO explains the decision in more detail.

Google documents ads above, below or inside AI Overviews, with different market eligibility. As checked on 30 September 2026, Austria is not listed for ads inside the overview; this does not rule out eligible placements above or below. Advertisers cannot directly select only the inside placement. Google Ads documentation.

Search, Shopping and Performance Max serve different tasks. PMax is not a compulsory remedy for organic losses. AI marketing workflows still need review and clear responsibilities. Meta Ads uses Facebook and Instagram; LinkedIn advertising is a separate platform. Compare channels by their job and economics.

How should you measure impressions, visits and commercial outcomes?

Search Console now documents a Generative AI performance report covering AI Overviews and AI Mode link impressions. Google dates worldwide rollout to 31 August 2026 while retaining caveats about report access and sufficient impressions. Check availability for the property. This is an impressions report, not a separate cost-per-lead calculation. Report documentation.

Google also documents Search generative AI control in Search Console settings. It allows inclusion or exclusion from specified generative features and may be inherited from a parent property. Eligibility checks should include the effective setting. Its scope differs from ordinary Search and training preferences. Control documentation.

Study ordinary clicks and CTR in search reporting, visits and actions in analytics, and qualification and deals in CRM. Keep manual answer observations separately: question, response, date and source URLs. Define mention, recommendation and citation shares using the AI visibility check method.

An impression does not establish attention, and a link does not establish a click. A later branded search might follow exposure to an answer, but without supporting data that remains a hypothesis. Preserve unknown acquisition sources instead of assigning all revenue to new content.

Three actions for this week

1. Record the starting point. Select valuable URLs and queries. Compare impressions, clicks and leads over comparable periods, and retain observed overviews by market and device.
2. Improve one to three pages. Clarify the answer, conditions, author and evidence. Check the path to the service and contact; avoid a site-wide heading change without a specific reason.
3. Check access and repeat. Identify a confirmed technical problem, agree any restriction changes and rerun the same observations. Keep business metrics separate from technical checks.

This helps decide whether to improve content, correct facts or run a limited advertising test. It does not set an indexing deadline or promise more recommendations after one change.

Frequently Asked Questions

Will AI Overviews stop all organic traffic?

There is no basis for that conclusion across every website. Check your query group and commercial actions. Some readers need only a summary; others need details and selection criteria.

How can I improve the chance of my page being used?

Provide an accessible, useful answer with verifiable authorship, sources and conditions. Check applicable search inclusion settings. No universal setting obliges Google to include your link.

Can I limit use of my content in AI Overviews?

Yes. Search Console has a generative AI inclusion control. Page-level snippet and indexing controls have separate effects, and Googlebot must be able to read them. Choose the intended restriction and check its effect on other Search results before changing it.

Do B2B leads become better after an overview?

Measure this for your business. Compare qualification, processing cost and deals alongside visits. An assumption about user intent is not a promise of higher conversion.

Should I continue ordinary SEO?

Yes. Indexable content, useful information, navigation and working links remain important. Checking AI answers and their sources complements those tasks.

How can I measure presence in AI Overviews?

Use the available Search Console AI impressions report and retain answers for important queries. Do not treat a fall in overall CTR as the exact traffic share taken by overviews. Check enquiries and deals separately.

Is it still useful to publish definitions?

Yes, when they help buyers understand a service or documentation. Connect them to practical detail and conditions. Publishing hundreds of near-identical definitions merely for search volume is not a useful goal.

Is llms.txt mandatory?

No. Google Search ignores it for visibility and ranking. Other systems may use the format for their own tasks. Check important HTML content and actual access problems first.

For the page-level controls mentioned above, see Google’s robots meta specification. nosnippet, data-nosnippet, max-snippet and noindex have different scopes; they should not be treated as interchangeable.

Discussing the change for your website

Save affected queries, visits and enquiry figures. GEO diagnosis can define a testable barrier for one market, language and question set. For an initial snapshot, use the preliminary GEO check; it is a limited first step, not a complete free project.