Press review 19.11.2026: citations without clicks, bidding across the funnel and one tag container

In one study 85% of the pages AI retrieved never made it into an answer. Google merges Google tag and Tag Manager into one container, and Journey-Aware Bidding counts conversions your bid strategy does not show.

Share
Press review 19.11.2026: citations without clicks, bidding across the funnel and one tag container

AI fetches your page and does not cite it - 85% of pages were dropped in one study

AirOps tested 15,000 prompts and more than half a million retrieved pages. As many as 85% of those pages never appeared in an answer as a citation. Being retrieved is only the first step, not visibility.

Position on the retrieval list matters. The first retrieved page had a 58% chance of being cited, pages further down only 14%. Almost a third of cited pages showed up solely through a follow-up query, not through the original prompt.

On top of that, Semrush counted domain appearances: most were citations without any brand mention. You are in the answer, but your name is not.

What to do: move the answer to the question in your title into the first paragraph. If it sits on the third screen, the model will not use it.

Sources:

In pharma, ranking earns citations but brands drop out of AI answers

Research performance.io and Semrush analysed four years of US pharmaceutical queries. 84% of AI Overview citations came from pages ranking in the top ten. Ranking still matters.

Pharma brands held 9.2% of first positions but only 4.2% of citations. Google picked third-party and medical publishers more often than manufacturer pages.

For medical queries YouTube accounted for 1.8% of citations, though across the whole web it accounts for about a quarter. In regulated topics, generic AI visibility advice stops working.

What to do: compare the pages ranking for your priority queries with the sources inside AI Overviews and check who Google picks instead of you.

Sources:

Your brand on someone else's page works faster than your own domain

Barnacle SEO means getting your brand onto pages that already rank: directories, review portals, forums and Reddit threads. You do not build domain authority, you borrow it from a page Google already trusts.

Ahrefs says it appears in AI answers 89% of the time thanks to mentions on other sites. A study of 10,000 brands found every mention type correlated with presence in ChatGPT and AI Overviews, and in an analysis of 260 million prompts Reddit threads had the strongest link.

The line is sharp. In 2024 Google devalued coupon sections on CNN and USA Today as site reputation abuse. Paid placement is safe when it is disclosed and the link is marked as sponsored.

What to do: pick three pages ranking for your hardest keywords and check whether your brand is mentioned there.

Sources:

More than half of Search Console impressions are anonymous - real questions live in sales calls

Suganthan Mohanadasan checked his own data: 57.7% of Search Console impressions landed in the anonymised pool over 59 days. Long, conversational questions are undercounted in GSC.

The generative AI report in Search Console shows URLs appearing in AI Overviews and AI Mode, but carries no query data. You cannot reconstruct what users asked from it.

The alternative is recordings of sales calls, discovery calls, tickets and chats. A question from a real conversation already looks like a brief: platform, timeline, two possible causes and a decision to make.

What to do: rewrite questions from your last ten client calls the way someone would ask an AI assistant, and count repeats. That is your demand data.

Sources:

AI Overviews take 39.8% of organic clicks while paid clicks hold

In a field experiment the arrival of an AI Overview cut outbound organic clicks by 39.8% while paid clicks stayed flat. A new meta-analysis pulls those measurements together in one place.

Position-one CTR with an AI Overview is 1.6% against 3.7% without it. Traffic is smaller but not gone: in Wynter's study 72.5% of people visited vendor sites after an AI chat.

AI Overviews now cover most commercial queries in the US, and consumers hand AI the choice only at the start of a purchase, not at the decision.

What to do: report organic clicks, AI mentions and citations, and leads separately. A single visibility number hides what is actually happening.

Sources:

Journey-Aware Bidding optimises on conversions your strategy does not show

Journey-Aware Bidding can use secondary conversions as a bidding signal even when they are excluded from your bid strategy. It only works with Target CPA and Maximize Conversions.

It is in beta and the interface does not tell you whether your account is included. Assume it is. The feature ignores conversion actions in the Other category, so the layout of your goals now matters more.

At one client with JAB enabled, scheduled meetings rose 57% and MQLs 52%. B2B sales cycles of 12-18 months fall outside Google's 90-day conversion window, so later-funnel signals are the only ones the model can see.

What to do: move actions you do not want optimised - disqualified leads for example - into the Other bucket and mark them as secondary.

Sources:

Microsoft rebuilt its bidding models for the first time in a decade - the change arrives unannounced

Microsoft Advertising is rolling out new automated bidding models and calls it its biggest update in a decade. The upgrade is applied automatically as it reaches more accounts, so campaigns may start spending differently before you react.

In value-based strategies Microsoft reports 10% more conversion value per click at a similar ROAS. Moving from Target CPA to Target ROAS delivered 8% more value, and from Enhanced CPC to Maximize Conversions 18% more conversions.

Bids react to target changes almost twice as fast and budgets pace more evenly through the day. Your own max CPC caps can block that flexibility.

What to do: review campaigns with max CPC set and change CPA/ROAS targets in steps of about 15%, reviewing after one conversion cycle.

Sources:

Google tag and Tag Manager are now one container - migration in January 2027

Every Google tag will become a Tag Manager container, with no separate GTM install. Sites running only a Google tag will gain interface-driven tagging, debugging tools and version control.

New Google tag and GTM snippets share the same format and no longer include the gtag config command. Container permissions will also control access to data in Tag Assistant.

Google says existing tags will behave as they do today, so you do not need to rebuild tracking. Unsupported installations will be upgraded in January 2027.

What to do: check whether your setup is on the unsupported list and inventory snippets containing gtag config before Google gets there.

Sources:

Google Ads calculates confidence intervals and statistical significance with jackknife resampling. The method discards single extreme conversions that would otherwise produce a false positive.

In templated tests traffic is split 50/50 inside a single campaign rather than through a campaign copy. The test arm learns faster and the test covers one change at a time.

A pre-post comparison, meaning before and after, does not control for seasonality or anything else. Without statistical significance the gap between arms may be plain noise.

What to do: when you change your bidding strategy, run an experiment instead of comparing months. One lever at a time, or you cannot attribute the effect.

Sources:

AI, tools and automation

Perplexity is not worth optimising for, and ChatGPT drives 80% of agentic traffic

According to Pawel from Paq Studio, Perplexity is a marginal player today. Its boom came when packages were bundled with mobile carrier subscriptions, not from a real search advantage.

ChatGPT accounts for 80% of agentic traffic, but against Google's volumes that is still a few percent. ChatGPT and Claude tend to be overrated when what you count is traffic to a site.

That does not make the topic worthless. Being present in AI answers builds brand recognition even when clicks are few.

What to do: do not split budget into a separate channel for every chatbot. Watch Google and ChatGPT, and treat Perplexity as an extra.

Sources:

GPT-6.2 Astra is terser by default, and in the API editing history drops reasoning items

OpenAI published prompting guidance for GPT-6.2 Astra and Sol, both launched on 17 November. Astra writes shorter answers by default, so instructions asking it to be concise may already be redundant.

The API drops reasoning items when the client edits earlier history, switches to an older model version or changes organisation. That lowers answer quality and burns tokens to redo previous work.

Instead of editing an earlier message, append a new one telling the model to ignore older instructions. The model likes lists and tables, and waits for confirmation when intent is unclear.

What to do: review prompts in your automations for conciseness instructions and for history edits. Those are the two changes that most often break results after a migration.

Sources:

Marketing and content

The first sentence should be the answer, not a promise

The BLUF rule - bottom line up front - is back with AI search. The reader should get the answer immediately, not a promise that it lives somewhere further down.

A sentence saying you will learn how to run an audit is only a teaser. The answer is the first audit step, stated plainly, with details below.

The same pattern works for models. A passage retrieved from the top of the page is more likely to make it into an answer than a paragraph from the third screen.

What to do: open your latest article and mark the first sentence that actually answers the question in the title. Count how much text comes before it.

Sources:

AI visibility is 70% a marketing problem, not a prompt problem

A Search Engine Land author splits AI visibility into five layers: product and price, positioning and proof, presence, page readability, and only then the AI layer. Each one works only if the layer below is healthy.

The suggested split is about 70% of effort on the three lower marketing layers, 20% on SEO and 10% on GEO itself. Many plans are upside down, and then nothing works.

A simple product-layer test: ask a model on a logged-out account whether the brand is any good and how it compares with a competitor. A lukewarm answer is a product or price problem, not a prompt problem.

What to do: find the lowest broken layer and start there. Adding AI-optimised content will not help if the offer and the proof do not hold up.

Sources: