Google Search Console and Bing AI Performance: What the Reports Measure Since 31 August 2026 – and Why Google Hides Clicks

On 31 August 2026, Google finished the global rollout of its dedicated Generative AI report in Google Search Console. Four days later, corporate dashboards look different: anyone managing a B2B website can now open their performance tab and filter specifically for visibility inside AI Overviews, the conversational AI Mode, and generative cards in Discover.

Microsoft introduced its AI Performance report in Bing Webmaster Tools as a public preview back in February 2026 and extended it on 16 June 2026. On the surface, the central visibility debate of 2026 appears settled: the two largest search providers in the Western hemisphere now provide first-party reporting on how often your brand is read, cited, and used by AI models.

In reality, looking into the raw numbers exposes a structural asymmetry. What Google shows is carefully curated; what Google withholds is precisely the data that would reveal the financial cost of zero-click answers. Bing delivers genuine transparency, but on an audience fraction. Understanding how to read both consoles alongside your own web server logs is the only way a mid-sized enterprise can make informed decisions about AI visibility without wasting five-figure annual budgets on speculative vanity tooling.

The Asymmetry: What Google Shows vs. What Bing Delivers

To evaluate the new consoles, one must look at what parameters each platform actually records in its database. The differences are architectural, not aesthetic.

Why Google Conceals Clicks: The Economic Calculus of Zero-Click

Google’s developer announcement of June 2026, which took until the end of August to reach all verified properties, gives no reason for the missing clicks and click-through rates; it only says that additional metrics will be added over time. The help page for the report does not mention clicks at all.

In our view, privacy would be a fragile explanation: Search Console has filtered low-volume queries with privacy thresholds for a decade.

In our assessment, the motive is economic. Independent measurements show the size of the loss: Ahrefs found about 34.5 percent fewer clicks on position one where an AI Overview appeared, measured across 300,000 keywords in 2025, and about 58 percent in its re-run with December 2025 data. In the Pew Research Center data (68,879 searches, March 2025), a traditional result was clicked in 8 percent of visits with an AI summary and in 15 percent without. Seer Interactive saw the organic click-through rate on informational search terms with AI Overviews fall by 61 percent between June 2024 and September 2025. A link inside the summary itself was clicked in only 1 percent of visits to pages with a summary (Pew).

If Google isolated these clicks inside the Search Console interface, every CFO and CMO would see an arithmetic discrepancy: 100,000 generative impressions generating fewer than 900 visits. By omitting clicks and CTR, Google provides the illusion of engagement while masking the severity of traffic deflection.

What GSC Generative AI Impressions Actually Mean in B2B

When your Search Console shows 45,000 generative AI impressions on a technical landing page with no noticeable uptick in incoming leads, executive teams frequently jump to one of two wrong conclusions: either the data is fictional, or the website has failed.

Neither is true. In the German mid-market, generative impressions reflect three distinct states:

• Informational Saturation (Zero-Click Answer): If a buyer searches for statutory deadlines, ISO standards, or tax definitions, Gemini parses your page, synthesizes the conclusion into three sentences, and presents your URL in the secondary source carousel. The buyer’s intent is fully satisfied in the SERP. The impression is valid; the click is impossible.
• Entity Grounding (Brand Mention Without Direct Link Click): When an AI Overview compares providers in your sector, it frequently cites your product page to extract pricing parameters or capability specifications. The user notes your firm as an established player and initiates an unbranded evaluation later, or visits your home domain directly via navigation.
• Transactional Shortlisting: On complex queries ("specialist machine tool manufacturer Baden-Württemberg automation 5-axis"), the summary leaves nuances open. Here, citation links are clicked. This is the only segment that converts into immediate site traffic.

How to Measure True AI Traffic Without 800-Euro SaaS Dashboards

Since late 2025, an aggressive market for "GEO monitoring platforms" has sprung up. Most cost between €600 and €1,500 per month. Under the hood, nearly all of them do the exact same thing: they run a list of canned prompts through OpenAI or Perplexity APIs and count how many times your brand string appears.

That is a vanity metric. It measures how an LLM responds to an artificial prompt in an isolated API call, not how real buyers find your business. You can build a vastly superior, grounded measurement system in three steps using infrastructure you already own:

1. Server-Side Referrer & Log Filtering

Real AI-driven traffic leaves signatures. In Google Analytics 4 (GA4) or Matomo, configure a dedicated traffic segment filtering for the following source domains:

• chatgpt.com and android-app://com.openai.chatgpt (ChatGPT Web & App)
• perplexity.ai (Perplexity AI Search)
• claude.ai (Anthropic Claude Artifacts & Web links)
• copilot.microsoft.com and edgeservices.bing.com (Microsoft Copilot)

In your Cloudflare or Nginx access logs, verify these sessions. In our analyses, visitors arriving via generative engines spend an average of 2.8 times longer on page than organic searchers and view 40 percent more documentation pages. They arrive pre-qualified because the conversational model already resolved introductory context.

2. Calculate Your AI Overview Exposure Rate

Open Google Search Console. Open the separate Generative AI performance report (available to all sites since the rollout completed on 31 August); it is not a search-type filter in the regular performance report. Sum the impressions for your top 20 foundational service and insight pages over the last 30 days. Divide this number by the total traditional impressions of those same pages.

In our client audits across mechanical engineering, IT services, and legal advisory, this rate typically sits between 32% and 54% for technical informational queries. If your rate is below 15%, in our experience your pages usually lack structured entity signals, clear definitions, or concise table summaries that Gemini can extract.

3. First-Party Inquiry Attribution

Because native desktop apps (like ChatGPT Desktop or Claude CLI) and privacy-oriented browsers strip HTTP referrers, part of the generative visits lands in your analytics as "Direct / None".

The simplest and most reliable correction is non-technical: add an un-gated, open-text field to your contact or meeting booking form: "How did you hear about us or research our firm?". When prospects write "Asked ChatGPT for recommendations on industrial AI integration" or "Perplexity cited your article on CRA reporting", you have captured pipeline causality that no cookie or tracking pixel can replicate.

The Technical Architecture: IndexNow and Semantic Machine Data

If you want your website to be cited accurately when generative engines answer queries, two technical requirements must be satisfied:

• Real-Time Submission via IndexNow: Copilot does not crawl the entire web constantly; it ingests fresh material through Bing’s fast index. The IndexNow protocol pushes updated URLs directly to Microsoft Bing, Yandex, and Naver within seconds of deployment.
• Semantic Machine Schemas and llms.txt: LLMs do not read design; they ingest semantic structure. Implementing validated JSON-LD schemas (TechArticle, FAQPage, Organization) alongside an /llms.txt file provides AI crawlers with structured context without token bloat. This does not apply to Google Search: according to Google’s guide to optimizing for generative AI search (as of July 2026), Google Search does not use llms.txt files, and structured data isn’t required for generative AI search.

Conclusion

The arrival of Google’s Generative AI report on 31 August 2026 and Bing’s AI Performance console does not mark the end of SEO; it marks its formal bifurcation. Visibility in organic search can no longer be evaluated by blue links and raw click counts alone.

Google offers volume without accountability; Bing offers precision without scale. By combining Google’s exposure rates, Bing’s grounding queries, and your own server referrer logs, your business can navigate the generative shift with clarity rather than guesswork. If you want to audit your company’s actual AI footprint across these consoles, that is precisely the work we do in an initial consultation.

Frequently Asked Questions on Search Console, Bing AI Performance, and GEO Metrics