Press review

Press review 10.11.2026: Q3 numbers in Google Ads, French AI Overviews and AI cards for assets

Paid clicks on Google grew 8 percent, down from 13, and an analysis of more than 400 accounts found no trace of longer AI Mode queries in search terms reports. In France, AI Overviews cost one site half its clicks while others felt nothing.

Bartosz Bielawski

Bartosz Bielawski

· 7 min read
Share
Press review 10.11.2026: Q3 numbers in Google Ads, French AI Overviews and AI cards for assets

AI Overviews took half the clicks from some sites and none from others

AI Overviews went live in France in July, and some sites lost Google clicks right away. One site fell from about 750 to 400 clicks a day while its impressions stayed flat.

Sites about health and money lost more. One lost almost 70 percent of its clicks, and another saw its click-through rate fall from 1.2 to 0.4 percent. Two of the five sites studied saw no change at all.

So a drop in traffic alone does not tell you whether AI Overviews are to blame. You have to look at clicks broken down by page type and query group, because impressions can stay flat while clicks melt away.

What to do: break your Search Console report down by page template and check where clicks fell while impressions held steady.

Sources:

Moving from WordPress to a headless setup, meaning a CMS plus Next.js, can still wreck your visibility even when every URL stays the same. Plugins and the theme no longer handle SEO, and CMS fields, front-end components and routing logic take over.

The hardest failures to spot are missing canonicals, navigation links a crawler cannot follow, content moved into JavaScript, and removed pages that return a 200 instead of a 404. Losing structured data costs you rich results even when rankings do not move.

What to do: before launch, take a baseline on a sample of URLs - HTTP status, canonicals, structured data and internal links - and repeat it on production, not on staging.

Sources:

An SEO A/B test splits pages, not people - otherwise you are measuring cloaking

Splitting users tells you nothing about rankings, because Google sees only one version of a URL at a time, and showing the crawler a different version than users see is cloaking. So the test splits groups of pages built on the same template: half get the variant, half serve as the control.

The rule of thumb is that the section you test should get around 1,000 organic visits a day. In an example with 400 product pages, clicks rose 14.2 percent after six weeks, with a confidence interval of 8 to 19 percent.

What to do: write down the hypothesis, primary metric and decision rule before you start, and report the interval rather than a single number.

Sources:

Reddit and YouTube are showing up in AI answers, and 68 percent of searches end without a click

A review of AI surfaces shows where models get their citations. Vendor pages, integrators and tool directories account for between 52 and 72 percent of citations depending on the model, and review sites add a smaller share on top.

Reddit appears in 52 percent of AI Mode answers, and YouTube is cited far more often by Google surfaces than by Gemini. On Google, 68 percent of US searches between January and April ended without a click.

What to do: add profiles on directories and review sites to your visibility plan, alongside your own site and blog.

Sources:

The search terms report will not show you AI traffic - even across 400 accounts

An analysis of more than 400 Google Ads accounts in three markets found no trend toward longer queries in search terms reports. Google says AI Mode queries are three times longer, but the data does not show it.

The reason is textbook: Google leaves out terms without enough query activity, and long one-off questions rarely clear that bar. For AI Mode, the reported term is only an approximation of intent, not the actual query. Unreported terms hide about 40 percent of the volume, and that share has been flat for nearly three years.

What to do: do not rely on the search terms report to detect AI traffic - estimate the gap between report totals and ad group totals, and measure AI through the Search Generative AI report in Search Console.

Sources:

AI Mode ads now appear on second and third questions too

Google has started showing ads on follow-up questions within a single AI Mode conversation in the US. An ad can appear under the second, third and fourth answer, not just the first.

The reason is simple: AI Mode ads monetize less well than classic results, so Google shows more of them. For you, that means new placements inside the same campaign, with no setting to change.

What to do: if you run US accounts, check whether ads appear on follow-ups, and do not compare their click rates directly with classic search.

Sources:

Google admits AI Mode ads earn less, and paid clicks are growing more slowly

Paid clicks on Google Search grew 8 percent year over year in the third quarter, after 13 percent in each of the two previous quarters. Cost per click rose 4 percent.

Google says outright that monetization in AI Mode is lower, and that people now ask questions earlier, further from the purchase. Only campaigns with broad match or keywordless targeting - AI Max, Performance Max, Shopping and dynamic search ads - can appear in AI Overviews.

What to do: give AI Mode and AI Overviews a separate test budget with its own success metric, instead of judging them by the same cost per conversion as classic search.

Sources:

Google is testing weekly AI cards about your ad assets in Google Ads

Some accounts now show a beta page called Asset activity with weekly cards about creative assets. The cards can be filtered into four categories: featured, asset performance, asset diversity and industry trends.

They suggest new headline themes, point to the best-performing headlines and report rising queries on a topic. Google itself labels asset-level metrics as directional only. So far, the feature has no help page and no announcement.

What to do: check the Assets section, and test any suggested headline in your own asset group instead of rolling it out right away.

Sources:

Clicks get more expensive while ROAS falls for a second quarter in a row - right before peak season

A report built on 20,000 accounts and 250,000 campaigns shows the same pattern: click-through rate rose in every segment, but return on ad spend fell for the second consecutive quarter. For the first time, it includes country benchmarks for five markets.

The authors also compare Standard Shopping with Performance Max Retail and show the cost gap that has opened up between lead gen and e-commerce. A single shared CPA benchmark for both account types is therefore misleading.

What to do: put your own numbers next to the benchmark for your segment and country before you allocate budget for Black Friday.

Sources:

Analytics and measurement

Engagement gives an error bar eight times narrower than conversions

With a thousand clicks and a 1 percent conversion rate, the margin of error is about 62 percent. The same thousand clicks at 40 percent engagement give about 8 percent - an interval eight times narrower.

On one SaaS account, a demo request was valued at 41,822 dollars and a content download at 70 dollars. That gap shows why adding up every rung of the ladder inflates campaign value and breaks bidding.

A holdout test on brand campaigns confirmed that some value is created beyond the last click: when ads went dark in part of the country, engaged sessions and form fills rose against the forecast.

What to do: report on engagement, but bid on actions a customer has to take deliberately - otherwise the algorithm buys the cheapest empty visits.

Sources:

AI already handles 76 billion sessions a month, yet sends under 1 percent of site traffic

Six AI assistants logged 76 billion sessions a month worldwide in September, 37 percent more than a year earlier. Even so, AI accounts for less than 1 percent of traffic to websites on average.

The report's authors stopped comparing AI with search engines, because Google search can no longer be counted reliably. In July, ChatGPT drove 60 percent of AI referral traffic, and Claude posted the biggest year-over-year growth.

What to do: before you write "AI is half of search" into a channel plan, check how the report counts Google - otherwise your strategy rests on the wrong denominator.

Sources:

Marketing and content

Five working traffic sources instead of one - a 2027 plan from Szymon Negacz

The list of reasons companies lose customers starts with ads: fewer clicks on Google, bigger budgets and worse results. Then come reviews shown only on your own site, too many sales messages, and research a buyer does in ChatGPT instead of on a call with sales.

On top of that comes a bigger buying committee, where each decision maker has an AI assistant, plus FOMU - fear of messing up, which freezes the decision. A single document that speaks to the whole committee at once raises group agreement by 56 percent.

What to do: count how many traffic sources actually work for you, and plan to stay in touch with a buyer for a year, not just until the next campaign.

Sources:

In email, 20 percent of the effort goes to new ideas and 80 percent to repeating what works

A good email campaign carries one value with many reasons, not many values at once. Unsubscribing is easy, because making it hard damages both the relationship and the data.

The metric that matters is lifetime value against acquisition cost, not open rate. A/B tests run continuously on 10 to 20 percent of the list, and the rest goes out as the control group.

What to do: pick one topic per email, keep a standing test running on a slice of the list and stop judging campaigns by open rate.

Sources:

Organic content nurtures leads, it does not generate them - and that changes your publishing plan

In the first two stages of a content machine, the goal is nurturing leads, not generating them. Buyers usually arrive through another channel, and the content only confirms that the company is credible and real.

Only later comes publishing on every platform and maxing out each one's capacity. That is when content starts capturing demand instead of just helping close a decision.

What to do: work out which stage you are in, and do not judge content by leads when its job is nurturing.

Sources:

Read more