The Same AI Rollout Produced Different Results

The Visibility Report #35 | Week of September 15-21, 2026

France now gives us two useful, conflicting pictures of what happened after Google launched AI Overviews there. Ahrefs found a sharp click-through-rate decline among the sites most exposed to the feature. Le Monde says its total audience did not suddenly fall.

Those findings can both be true. One measures a selected cohort's Google Search performance. The other describes one publisher's total audience, supported by app use and subscriptions. The useful question is not whether AI Overviews always destroy traffic. It is which pages, queries, channels, and business outcomes changed for your site.

Ahrefs found a 23.1% CTR drop in its most-exposed French cohort

Ahrefs data scientist Xibeijia Guan analyzed 963 French domains in Google Search Console across the 28 days before the July 22 launch and the nine days after it. The group most exposed to AI Overviews lost 23.1% of its click-through rate. Ahrefs also reports that almost half of the sites in its panel lost more than 10% of daily clicks.

This is a strong warning, not a universal forecast. The panel covers a short post-launch window and does not report conversions, direct visits, app traffic, subscriptions, or every French site. Segment your own Search Console data by pages and query groups with likely AI Overview exposure before applying the benchmark.

Source: Ahrefs​

Le Monde reports no sudden audience drop

Le Monde Group CEO Louis Dreyfus says the publisher did not see an abrupt decline in total audience after AI Overviews launched in France. He points to app traffic, international readership, and subscriber growth. The French press alliance APIG also says early reports did not show a massive total-audience impact, even as it cites an Arcom estimate of a 33% to 38% loss in Google Search traffic in other European markets.

Le Monde did not publish a clean Google Search referral breakdown, so this does not rebut the Ahrefs study. It shows why channel mix matters. A business with strong direct, app, email, or subscriber relationships can absorb a search decline differently from one that depends on informational clicks.

Source: Search Engine Journal​

Traffic and conversions are separating

Search Engine Journal editor-in-chief Katie Morton reports that 39% of surveyed search professionals saw traffic decline or remain flat while leads and conversions held or improved. Only 15% lost both traffic and conversions. The survey does not prove that AI answers caused the split, and publishers still need audience scale. It does show why traffic alone is an incomplete business measure.

Put page-level traffic beside lead quality, assisted conversions, pipeline, revenue, and branded demand. If a lost informational visit never had commercial intent, it is not equivalent to a lost buyer. If the top of the funnel disappears, the damage may arrive later. Measure both.

Source: Search Engine Journal State of Search 2027​

Google is testing payments for content used in AI answers

Google confirmed an early pilot that pays participating sites when their content contributes to answers in Gemini, AI Overviews, and AI Mode. Search Engine Journal's Matt Southern, citing Digiday, reports that at least dozens of publishers have been approached. Google has not disclosed eligibility, rates, attribution rules, or whether the program will expand.

This is publisher compensation for source material, not paid placement inside an answer. Maintain a page-level record of AI citations, referrals, licensing inquiries, and content costs. If a broader program appears, you will need evidence to decide whether its terms cover the value and expense of the material being used.

Source: Search Engine Journal​

Cloudflare separates AI training from search access

Technical consultant Slobodan Manic reviewed Cloudflare's announced Bot Preference Sync, which turns three dashboard settings into directives added to a site's robots.txt file. His concern is practical: category-level settings cannot express every per-crawler policy, and the generated file can differ from enforcement at the edge.

A site may want to block model training while allowing systems that retrieve pages for search answers. Robots.txt remains a request rather than enforcement, and the product's documentation was incomplete when Manic reviewed it. Write down your policies for training, search retrieval, and ordinary indexing separately, then test what each user-agent receives.

Source: Slobodan Manic in Search Engine Journal​

A retrieval test can separate access problems from ranking problems

Torque Partnership technical director Chris Green proposes a simple diagnostic for AI search: take a distinctive 20- to 30-word passage from a page, ask a search-enabled assistant for exact matches, and check whether it returns the correct URL. A match suggests the system can retrieve and attribute the page. It does not prove that the page will rank, earn a citation, or drive traffic.

That distinction makes the test useful. If the page cannot be retrieved, investigate discovery, crawling, rendering, or source access. If it can be retrieved but is not selected, move to relevance, passage quality, authority, and competitive sources instead of treating every miss as an indexing problem.

Source: Chris Green in Search Engine Journal​

Published AI policies can become search evidence

SEO-PR president Greg Jarboe argues that an AI accountability document can serve two purposes: govern internal use and give search systems a clear, indexable account of a brand's policies, reviewers, and data protections. His example comes from a school district rather than a commercial brand, so it is a template, not proof that policy pages earn citations.

The useful move is operational. Name who reviews AI output, what data may be used, how corrections work, and which claims are current. Publish the parts customers and answer engines need to evaluate. Governance that only exists in an internal slide deck cannot help a buyer or a retrieval system understand the policy.

Source: Greg Jarboe in Search Engine Journal​

From Practitioners

• ​Xibeijia Guan: The Ahrefs analysis provides a cohort-level exposure method worth recreating with your own Search Console data, without turning 23.1% into a universal benchmark.
• ​Louis Dreyfus: Le Monde's channel mix shows why total audience and Google Search referrals should not be treated as the same measure.
• ​Katie Morton: Traffic still measures reach, but the State of Search split makes conversions and pipeline necessary companions.
• ​Chris Green: Exact-passage retrieval testing can identify whether access is the first hurdle before teams rewrite content.
• ​Slobodan Manic: Crawler policy needs to match both robots.txt and actual edge enforcement.
• ​Greg Jarboe: A public accountability document can turn governance into evidence that customers and search systems can inspect.

From the Tool Blogs

• Semrush: An AI citation outreach workflow uses cited pages as the target list instead of starting with generic link prospects. Published September 21.
• Lumar: A GEO diagnostic checklist separates technical access, passage relevance, authority, and source selection. Published September 14.
• Profound: Command Center gives teams a shared operating view of AI visibility work. Published September 14.
• Profound: A $180 million Series D at a $1.8 billion valuation is a market signal, not evidence of product performance. Published September 16.
• Peec AI: An analysis of ChatGPT's Labrador index family argues that retrieval cannot always be reduced to scraped rankings from another search engine. Published September 4.
• Peec AI: Brand Perception separates mention visibility from the attributes, objections, and factual claims attached to a brand.

What to watch

• Whether Ahrefs' French CTR result holds over a longer window and across lower-exposure cohorts.
• Whether Le Monde or APIG publishes a Google Search referral breakdown beside total audience.
• Whether Google's publisher-payment pilot discloses eligibility, rates, or attribution rules.
• Whether Cloudflare's generated robots.txt and edge enforcement stay aligned across crawlers.
• Whether exact-passage retrieval tests remain stable across signed-in states, models, and search providers.

What to do next

• Build an AI Overview exposure segment instead of relying on a sitewide average.
• Put search clicks beside conversions, pipeline, and direct-audience measures.
• Run exact-passage retrieval checks before diagnosing a citation or ranking problem.
• Record AI citations and publisher-program inquiries at the page level.
• Audit crawler policy separately for training, search retrieval, and indexing.
• Publish the customer-facing parts of your AI accountability policy.

The headline number is useful. Your own exposure, channel mix, retrieval state, and business outcomes decide what it means.

The Visibility Report | Will Scott​
This newsletter is produced collaboratively by Will Scott and Bob, an AI agent. Human oversight, AI efficiency.
​Subscribe at thevisibilityreport.com​