Press review

Press review 12.11.2026: channel allocation in PMax, ChatGPT traffic missing from analytics and thousands of unindexed pages

ChatGPT accounted for 11.2% of visibility in analytics, yet 25% of customers named it as their source. Google Ads cut the learning period to seven days and stopped restarting it after a budget change.

Bartosz Bielawski

Bartosz Bielawski

· 8 min read
Share
Press review 12.11.2026: channel allocation in PMax, ChatGPT traffic missing from analytics and thousands of unindexed pages

Move to a CDN and Google will crawl more - but warm the cache first

When you move a site to a CDN, response times drop, Google reads that as a hardware upgrade and speeds up crawling. The catch is that a new CDN starts with an empty cache, so the first request for every URL goes straight to the origin server.

People usually warm the cache by visiting a handful of popular pages. A crawler does the opposite: it works through the entire inventory of URLs, so the load spike lands exactly when the origin is most exposed. The second trap is blanket bot protection with no exceptions. It blocks Googlebot too, and then pages and shopping listings start dropping out of search.

What to do: warm the cache before you switch DNS, add a rule in the CDN that lets verified bots through, and if the origin struggles, return 429 or 503 for a few hours, or two days at most.

Sources:

Before you ship thousands of pages, check how many actually get indexed

Programmatic SEO is a way of publishing, not a type of content. It has three parts: a query pattern, your own dataset and a template. If you cannot say where every number on the page comes from and when it was last updated, you do not have a programme - you have mass-produced text.

The most common mistake in this discipline is shipping several thousand pages before anyone knows the indexation rate. A safer path is a test batch of 25 to 50 pages, measurement from server logs, and a rule for removing pages that is written before launch rather than after an update. The context has shifted: since tightening its policy on scaled content, Google reports 45% less low-quality content in results, and 68% of US searches end without a click.

What to do: measure results per template, not per site. In a zero-click world, the goal is no longer traffic but presence in AI answers.

Sources:

Analytics shows ChatGPT, but customers name it twice as often

Marketing Miner compared what analytics sees with what customers say. In GA4, ChatGPT accounted for 11.2% of visibility, but 25% of customers named it as their source at purchase. The rest of that traffic disappears into direct and referral traffic, because AI assistants prefer to keep users on their own platforms.

The gap in quality is often bigger than the gap in volume. For Jim Beam, organic Google results converted at 0.06%, while AI traffic converted at 5%. That is why session counts are misleading: AI is still a marginal source of visits, but the people who arrive through it are much further along in their decision.

What to do: add one question - "How did you hear about us?" - to your checkout and forms, and compare the answers with your referral data. Without it, AI traffic stays invisible and the content budget for AI answers is hard to defend.

Sources:

The Learning status now lasts seven days at most, down from between seven and 14. Changes to budgets, bid targets, creatives and targeting no longer trigger it at all, and budget and target adjustments take effect immediately.

Learning is still triggered by three things: a new strategy, a change in the strategy's composition and a change of conversion goal. Performance Max and Demand Gen campaigns may now need up to seven days to start serving, two days longer than before. A missing status does not mean results are stable - Google only describes when it shows the label.

What to do: after a budget change, watch CPA, ROAS and spend pace instead of the strategy status. Update any internal rule that says not to touch a campaign for two weeks, because the assumption behind it is out of date.

Sources:

The new reach limit in Google Ads can leave a campaign with no audience

A setting called Limit audience reach has appeared in the Google Ads interface, with three options. The default, and the one Google recommends, is no limit at all. The second narrows reach to people matching only the age and gender segments, and the third to people who match all selected segments.

The strictest option is described as intended for remarketing, where your own data is the only segment. Google warns that limiting reach can hurt performance and leave a campaign with no eligible users. With the strictest setting, an extra warning appears: targeting will exclude people who may convert but do not look like the defined segments.

What to do: leave reach unlimited in campaigns aimed at new customers, and apply the restriction where you target your own lists. After the change, keep an eye on audience size, delivery and conversions.

Sources:

Channel priority in PMax will not buy pricier traffic, and a zero allocation shuts a channel down

Performance Max still treats Search, YouTube, Discover, Gmail and Maps as a single group of Google channels. You cannot set different priorities within that group, so there is no way to put Search ahead of YouTube within one budget. Priority does not override CPA and ROAS targets either - a campaign will not buy more expensive inventory just because you marked a channel as important.

Allocation percentages are calculated from the average daily budget, not from actual spend, so after a budget increase the same percentage means a bigger sum. Limits above zero work on a best-effort basis, and only a zero allocation blocks spending on a channel entirely. Search campaigns with Display enabled need a settings review before February 2027, when Google retires the current setup.

What to do: compare networks by the cost of a qualified lead and by margin, not by form-fill CPA. A bigger Search share of spend may simply mean that partner spend fell.

Sources:

A new parameter shows which Google Ads clicks came from AI Max

Google Ads has added a ValueTrack parameter called search_term_matchtype. It takes four values: e for exact, p for phrase, b for broad and a for AI Max expansions and keywordless matches. The letter a is the new part.

The parameter sits alongside the existing matchtype parameter. Matchtype describes the keyword's match type, while search_term_matchtype shows how the actual search term was matched. That means a click on an exact match keyword that carries the value a can be flagged as AI Max traffic.

What to do: add the parameter to your tracking template and build a separate AI Max segment in analytics. Bear in mind that the parameter alone will not measure AI Max's incremental value - that requires a test.

Sources:

AI and tools

ChatGPT mentions stores most often, but no brand appears in more than two percent of answers

Ahrefs counted how many of 5.5 million US ChatGPT answers mention each brand. Walmart comes first, appearing in 95,605 answers, or 1.73% of the total. Target, in second place, appears in half as many.

The structure of the answers says more than the ranking. Six of the top ten spots belong to stores, and carmakers take 15 places in the top hundred. Stores appear when people ask where to buy, while product brands appear when they ask what to buy and how to use it. Google almost never shows up in shopping answers.

What to do: measure your brand's share of answers against competitors, not just citations of your own pages. Also make a note of the questions where only competitors appear - that is your list of gaps to close with content.

Sources:

Memes are the smallest part of what people do with Google's image generator

Google analysed 2.1 million conversations with its Nano Banana image generator in Gemini and grouped them into five categories of use. The analysis covered only prompts and conversation structure, not the images themselves.

In the social and viral category, nearly 80% of images featured people, while memes and humour turned out to be the rarest. In the marketing category, the most common requests were for exact text in the image and detailed typography instructions. Users started over from scratch about twice as often as they asked for small changes.

What to do: treat the generator as a tool for product visuals and consistent branding, not for memes. In the prompt, specify the exact text, typography, layout and a list of forbidden elements, and start over more often instead of stacking up small fixes.

Sources:

Marketing, sales and business

A podcast will not save your funnel if the top is missing - you need five times the content

A training company stalled at a few dozen customers a year, even though it ran a podcast with 10,000 downloads a month. After bots were filtered out, it turned out that 500-600 people actually listen to each episode, and 5% of them convert.

A podcast and an email list sit in the middle and at the bottom of the funnel. What was missing was the top, the awareness layer, and that is why growth stalled. The owner was advised to increase output fivefold rather than double it, and to lead with video, because video is easier to turn into audio than the other way round.

What to do: replace the direct call to apply with a concrete lead magnet, such as the first module of the training. Then cut your best posts into short formats and stitch them together into longer YouTube videos.

Sources:

Fresh proof persuades more than the best case study from years ago

Effective social proof has four traits: it is recent, visual, abundant and tied to a pain the customer feels. A review from last week works better than a five-year-old case study, and a photo of the result works better than a sentence about it.

Most businesses have more proof than they think - they just never collect it. A hundred five-star reviews on a local platform is an average result today, but showing all of them at once makes an impression that a single review never will.

What to do: collect proof continuously and present it in a visual form. In customer stories, start with the problem rather than the outcome - that is when readers recognise themselves and keep reading.

Sources:

A service bundle closes better but eats into margin - raise the price, do not break it up

A services company sold its upmarket clients a bundle of all its services. Sales were good, but margins were lower and there was little room for add-ons. Without the bundle, it was the other way round: deals were harder to close, but margins were better and there was more room to expand the contract.

The advice in that situation was not to break up the bundle, but to raise its price. Among wealthier clients, a price difference of a few dozen percent often goes unnoticed, and margin rises immediately. The expert also pointed out that if sales are healthy, the offer is probably not what is holding the business back.

What to do: work out what the bundle gains you in close rate and what it costs you in margin, and test a price increase before you change the structure. Separately, look for the real constraint on growth, because it may not be the offer.

Sources:

Read more