> ## Content Index > Fetch the complete content index at: https://bielawski.pro/llms.txt > Use this file to discover other available public pages before exploring further. # Week 37/2026: ad strength in Google Ads, a bare-bones ChatGPT ads panel and labelling AI content - URL: https://bielawski.pro/en/week-37-2026-ad-strength-in-google-ads-a-bare-bones-chatgpt-ads-panel-and-labelling-ai-content/ - Published: 2026-09-13T17:00:00.000Z - Updated: 2026-11-05T14:52:10.000Z - Description: Ad strength in Google Ads says nothing about results: in a large study, campaigns rated “poor” did better than those rated “excellent”. On top of that, the ChatGPT ads panel does not yet let you exclude keywords. - Author: Bartosz Bielawski - Tags: SEO, Google Ads, AI, Marketing, #en ## The week in three sentences In advertising, this week busted the myth that a green indicator in the panel means a good campaign. Google keeps turning AI Overview into a full conversation, and AI happily draws on YouTube because it can read it. Finally, a legal reminder: AI-made content has to be labelled, while new models from OpenAI and Google are quickly changing what you can do with images and video. ## Ad strength in Google Ads does not tell you whether a campaign makes money *Signal: 5 independent sources this week, 8 in 30 days; tracked since 2021-08-02.* Ad strength in Google Ads only checks whether an ad has a full set of headlines, descriptions and images. It says nothing about whether the ad brings in sales. A study of 20,000 accounts found that campaigns rated “poor” performed better than those rated “excellent”, and the latter paid more per click (Optimizer). On top of that, automatic asset optimisation is on by default and can change your copy, your images and even your landing page URL. What to do: judge ads by conversions, not by the indicator, and check in the asset settings what Google changes on your behalf. **Sources:** - WhitePress, 2026-09-12: [Atak klonów w Meta i Google Ads. Pułapki AI i automatyzacji, które przepalają Twój budżet](https://www.youtube.com/live/IiExs1vEm2A&t=2093s?ref=bielawski.pro) ## You can buy ads in ChatGPT, but you cannot exclude keywords *New signal: one source so far.* The ChatGPT ads panel is very basic for now. The only exclusion you can set is location. There are no keyword exclusions, and audience targeting appears in the panel but cannot be switched on. So you cannot filter out queries that do not fit your offer or narrow a campaign to specific people. Your budget can easily leak into random conversations. What to do: start with a small budget and check results often before you spend more. **Sources:** - WhitePress, 2026-09-12: [Atak klonów w Meta i Google Ads. Pułapki AI i automatyzacji, które przepalają Twój budżet](https://www.youtube.com/live/IiExs1vEm2A&t=3322s?ref=bielawski.pro) ## AI Overview is turning into a full AI Mode answer *Signal: 2 independent sources this week, 12 in 30 days; tracked since 2024-07-24.* Google is testing an AI Overview that expands into the full AI Mode answer on its own, without a “Show more” click. A box for the next question sits right below it (Search Engine Roundtable). Users get the whole picture without visiting a site, so clicks on results may drop. At the same time, Google questions whether the impact of AI Overview can be read from overall search referral stats. What to do: compare traffic to individual pages before and after AI Overview appears, not just the site-wide total. **Sources:** - Search Engine Roundtable, 2026-09-12: [Google Testing Making AI Overviews Into AI Mode Responses](https://www.seroundtable.com/google-ai-overviews-push-ai-mode-responses-41974.html?ref=bielawski.pro) - Search Engine Journal, 2026-09-12: [What Wikipedia Reveals About AI Overviews And Web Traffic](https://www.searchenginejournal.com/what-wikipedia-reveals-about-ai-overviews-and-web-traffic/589042/?ref=bielawski.pro) ## AI-made content has to be labelled *Signal: 3 independent sources this week, 7 in 30 days; tracked since 2026-07-25.* A new part of the AI Act, the EU law on artificial intelligence, has applied since 2 August. Its Article 50 requires that people know when they are talking to AI or looking at AI-generated content. That covers marketing too: images, video and text published for customers. Some video tools already add a note on their own saying the material was made with AI. What to do: agree on a simple company rule for how and where you label AI content, before anyone asks. **Sources:** - WhitePress® Global, 2026-09-12: [Beyond AI SEO: Publishers Can Win The Attention Game](https://www.youtube.com/watch?v=e9eUG4dYQns&t=2477s&ref=bielawski.pro) - WhitePress, 2026-09-12: [Atak klonów w Meta i Google Ads. Pułapki AI i automatyzacji, które przepalają Twój budżet](https://www.youtube.com/live/IiExs1vEm2A&t=1411s?ref=bielawski.pro) - Michał Barczak, 2026-09-08: [Meta POZWAŁA OpenAI, AI Act ZACZĄŁ obowiązywać - co to zmienia](https://www.youtube.com/watch?v=C7QwN%5Fr5%5FHA&t=56s&ref=bielawski.pro) ## AI reaches for YouTube because it can read it *Signal: 2 independent sources this week, 2 in 30 days; tracked since 2026-09-12.* AI answers often cite YouTube videos. The reason is simple: every video has an automatic transcript, a written version that AI reads like an article. YouTube is also full of videos that explain things step by step, plus plenty of structured data about each video (NoGood). For AI, that makes it a ready-made, easy-to-understand source. What to do: record short explainer videos on your key topics and fix the automatic captions. **Sources:** - NoGood, 2026-09-12: [How to Get Your Videos Cited by AI](https://nogood.io/blog/youtube-aeo-guide/?ref=bielawski.pro) - WhitePress® Global, 2026-09-12: [Feed the machine: How to get into AI answers with brand mentions](https://www.youtube.com/watch?v=7G6r3G1LbZY&t=1013s&ref=bielawski.pro) ## Astra uses fewer tokens, Fable writes better code *Signal: 2 independent sources this week, 2 in 30 days; tracked since 2026-08-21.* After a week of testing, the difference between the two new models is clearer. GPT-6 Astra uses almost three times fewer tokens (the units of text you pay for) than Fable 5.1 (Theo). Fable 5.1, on the other hand, writes code that is easier to merge, while Astra is clearly better at operating a computer. Fable in Claude Code also has a much larger context window, so it can hold more material at once. What to do: pick a model for each task rather than one for everything. Give long documents and code to Fable, and work inside apps to Astra. **Sources:** - AI LABS, 2026-09-12: [It's Over... This Just Ended The GPT 6 Astra Vs Fable 5.1 Debate](https://www.youtube.com/watch?v=BwNNcob3-A0&t=156s&ref=bielawski.pro) - Theo - t3․gg, 2026-09-12: [Fable Vs Astra Debate Is Over](https://www.youtube.com/watch?v=P7bxbDSnZRM&t=3738s&ref=bielawski.pro) - Theo - t3․gg, 2026-09-12: [Anthropic employees are quitting due to safety concerns](https://www.youtube.com/watch?v=4B4R2T4w7Kg&t=1111s&ref=bielawski.pro) ## ChatGPT's new image generator fixes only what you point at *Signal: 3 independent sources this week, 4 in 30 days; tracked since 2026-07-08.* OpenAI has released ChatGPT Images 2.5\. The biggest change is that you can edit a small part of an image without touching the rest. Images also have sharper detail and more natural light, and they stick closer to reference photos. The model fits more information into a single graphic, though testers also mention some surprising weaknesses (Futurepedia). What to do: for ad and social graphics, fix individual elements instead of regenerating the whole image, but always check any text in it. **Sources:** - Goldie SEO Tips, 2026-09-13: [ChatGPT Images 2.5 Just Became THE KING!](https://www.youtube.com/watch?v=2jrF0QbIbqQ&t=0s&ref=bielawski.pro) - The AI Advantage, 2026-09-12: [ChatGPT Images 2.5 Is Here. Together With Astra It’s Crazy](https://www.youtube.com/watch?v=dZnsz2RYoAQ&t=69s&ref=bielawski.pro) - Futurepedia, 2026-09-12: [ChatGPT Image 2.5 Is The New King (Fully Tested)](https://www.youtube.com/watch?v=bbmgWQKZ5ZU&t=0s&ref=bielawski.pro) ## Gemini shows you the moment viewers leave your video *New signal: one source so far.* Google has released Gemini 3.8 Flash, a fast model that breaks a task into steps and checks its own work before answering. It can also watch a video and point to the spots where viewers probably skipped ahead. It catches slow pacing, repetition and weak openings (Goldie SEO Tips). That job used to require expensive, slow models or a very patient human. What to do: run your videos through it before publishing and fix the first few seconds. **Sources:** - Goldie SEO Tips, 2026-09-12: [How To Rank On Google With Gemini 3.8](https://www.youtube.com/watch?v=30eDOVsePB8&t=56s&ref=bielawski.pro) - Goldie SEO Tips, 2026-09-08: [Google Just Gave Gemini SUPERPOWERS! 🚀](https://www.youtube.com/watch?v=xMnSdGlGnf4&t=47s&ref=bielawski.pro)