Week 36/2026: AI Mode instead of results, two new models and attribution in Google Ads
AI Overview now leads straight into a conversation with AI, and just over 1% of people click the source links. In the same week, OpenAI and Anthropic showed models that carry out tasks on a computer by themselves.
The week in three sentences
This week showed that a search result is turning from a list of links into the start of a conversation. Two new models, Astra from OpenAI and Fable 5.1 from Anthropic, shifted the line on what you can hand over to a machine end to end. In advertising, an old problem came back: the last click still steals credit from the channels that build demand.
Google pulls you into AI Mode before you can click a result
Signal: 2 independent sources this week, 5 in 30 days; tracked since 2026-07-08.
AI Overview is no longer a short summary with a “show more” button. Google confirmed that it now sends users from the overview into AI Mode by default, and the full answer loads straight away, along with a box for the next question (Edward Sturm).
For your site, this means the fight for a spot in the results more often ends without a visit. Just over 1% of users click the source links under the answer, so being cited won't bring traffic on its own.
What to do: check whether your key pages are cited, but treat it as brand visibility, not a traffic forecast.
Sources:
- Edward Sturm, 2026-09-04: Google Just Expanded AI Overviews: This Is What Still Gets Clicked
- Goldie SEO Tips, 2026-09-03: Apify AI Just Made Web Scraping Too EASY!
OpenAI's new model operates the computer instead of just writing about it
Signal: 3 independent sources this week, 3 in 30 days; tracked since 2026-09-05.
OpenAI showed GPT-6 Astra, a model built to operate a computer. It clicks, fills in fields and moves through programs on its own, instead of only suggesting what to do.
In tests it beats Fable 5.1, scoring almost 98% on maths tasks (Frontier Math), and it is cheaper per task. That is the difference that turns AI from an adviser into a doer.
What to do: give it one boring process you run every week and see how much is left to fix once it is done.
Sources:
- Theo - t3․gg, 2026-09-05: It's Here.
- Mikołaj Abramczuk, 2026-09-05: GPT-6 Astra: Wszystko, Co Musisz Wiedzieć (w 12 minut)
- The AI Advantage, 2026-09-05: GPT-6 Astra: 20 Real Examples From Useful to Almost Impossible
Fable 5.1 builds a whole project from one brief and costs less
Signal: 2 independent sources this week, 4 in 30 days; tracked since 2026-07-08.
Anthropic released Fable 5.1, also known as Claude 5.1. It is clearly cheaper than its predecessor, and the projects it delivers are of a higher standard.
In tests it built a spreadsheet with a valuation model, plus a presentation to go with it, from a single prompt. In another test it pulled assets from a given website and turned them into brand guidelines, author photo included. That is work you would normally spread over several days and several people.
What to do: hand it whole tasks from start to finish rather than single paragraphs, and always read the output before it goes to a client.
Sources:
- AI News & Strategy Daily | Nate B Jones, 2026-09-05: Claude Fable 5.1: Not Just Code. It Made Me A Film, 7 Sheets And 13 Slides.
- Jack Roberts, 2026-09-03: Fable 5.1 Just Solved AI Slop... I'm Done
Readers can pick your site as a preferred source in AI
New signal: one source so far.
Google extended Preferred Sources, the button for choosing favourite sources, to AI Overview, AI Mode, Discover and Top Stories. A selected site shows up higher in those places.
The user makes the call, so visibility goes to brands someone has already chosen. Your site then appears more often in AI answers, even when a competitor has the better article.
What to do: ask your regular readers to add your domain to their preferred sources, the same way you ask them to join your newsletter.
Sources:
- Edward Sturm, 2026-09-04: Google Just Expanded AI Overviews: This Is What Still Gets Clicked
You can now see whether a competitor beats you in ChatGPT with an ad or with content
New signal: one source so far.
SE Ranking started detecting paid placements in ChatGPT answers and in Google AI Mode. The visibility report now shows which spots are bought.
Until now, reports only showed where you stand and never explained why a competitor ranks higher. Without that, it is easy to rewrite pages that were never the problem. The difference matters in practice: a paid spot moves with budget, an organic one only with content and links.
What to do: before you rewrite a page, check in the report whether the competitor's lead is simply paid for.
Sources:
- Goldie SEO Tips, 2026-09-03: This SE Ranking Is Crazy!
Google Ads will count sales that come long after the click
New signal: one source so far.
Google Ads is introducing Qualified Future Conversions. The metric estimates sales that arrive up to 180 days after someone interacted with an ad.
Instead of a fixed 7- or 30-day window, Google looks for a mid-funnel signal, such as a branded search, and waits for the outcome. That surfaces sales that currently drop out of reports and look like a win for another channel.
What to do: compare the new numbers with the old ones on campaigns with a long decision cycle, and leave bids alone until you see the gap.
Sources:
- Google Ads, 2026-09-03: Attribution, incrementality & MMM: The modern measurement trifecta
The last click still steals credit from YouTube and TikTok
Signal: 1 independent source this week, 3 in 30 days; tracked since 2026-07-21.
Google recommends moving away from last-click attribution. Instead, it suggests data-driven attribution that also counts views.
Last-click attribution undervalues top-of-funnel channels like YouTube, TikTok and TV, because they rarely get the final click. Budget then goes where results are easy to count, not where demand is created. In practice, the channel that happened to close the sale gets the credit, even if the customer first came across the brand somewhere else entirely.
What to do: switch on attribution that includes views and confirm the difference with an incrementality test before you move money.
Sources:
- Google Ads, 2026-09-03: Attribution, incrementality & MMM: The modern measurement trifecta