> ## 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 47/2026: AI shopping, exact match under AI Max and two OpenAI models
- URL: https://bielawski.pro/en/week-47-2026-ai-shopping-exact-match-under-ai-max-and-two-openai-models/
- Published: 2026-11-22T08:03:08.000Z
- Updated: 2026-11-22T08:03:08.000Z
- Description: Six in ten Polish adults used AI to plan purchases in the past six months, yet only 11% would let it pick a product for them. In the same week, an independent analysis found Google Ads automation lifts conversion value by 13% while CPA rises by 16%.
- Author: Bartosz Bielawski
- Tags: Prasówka, #en

## The week in three sentences

AI moved into everyday shopping: people ask models about products, but they keep the decision. Google and Microsoft are rebuilding bidding so that advertisers control less of what they pay for. And SEO is learning that visibility in AI answers comes from brand mentions and correctly configured bots, not from the page alone.

## AI shopping is routine now, but AI-written copy is losing trust

*Signal: 3 independent sources this week, 4 in 30 days; tracked since 2026-07-21.*

Six in ten Polish adults used AI to plan purchases in the past six months (PromptEye). It is no longer an experiment.

Buyers who ask a model about a product hit 12 touchpoints with the brand, more than those who shop without AI (Ipsos). AI does not shorten the path to a purchase, it adds stops. The decision still belongs to the person: only 11% of consumers let AI pick the product for them (Gartner).

Your content is part of that journey, and trust in it is falling. Only 35% of Americans trust AI-generated content (Emplifi).

What to do: have a human check product copy written by a model, and track traffic from chats separately, or it disappears into referrals.

**Sources:**

- Search Engine Land, 2026-11-21: [Google's AI search advice sparks backlash from SEOs](https://searchengineland.com/google-ai-search-advice-seo-backlash-502610?ref=bielawski.pro)
- PromptEye, 2026-11-21: [Widoczność w AI — Przewodnik PromptEye v3.1](https://raporty.prompteye.com/api/e-commerce/2026/guideline?ref=bielawski.pro)
- Search Engine Land, 2026-11-19: [AI search meta-analysis: Q4 2026](https://searchengineland.com/ai-search-meta-analysis-q4-2026-502232?ref=bielawski.pro)

## An AI summary takes half your clicks, and almost nobody taps the link inside it

*Signal: 2 independent sources this week, 6 in 30 days; tracked since 2026-07-08.*

When an AI summary sits above the results, people click a regular link in 8% of visits. Without it, they do so in 15% (Pew).

The link inside the summary barely exists: it leads to a click in 1% of visits. A citation in AI is not a promise of traffic, and being in the answer is not enough for anyone to reach you. On queries with AI Overviews, organic CTR has already fallen to 0.61% (Seer Interactive).

What to do: report queries with and without an AI summary separately, and measure click loss by page type, not for the whole domain.

**Sources:**

- Search Engine Land, 2026-11-21: [Google's AI search advice sparks backlash from SEOs](https://searchengineland.com/google-ai-search-advice-seo-backlash-502610?ref=bielawski.pro)
- Search Engine Journal, 2026-11-21: [AI Citations: What To Check Before Reacting To A Lost One](https://www.searchenginejournal.com/ai-citations-what-to-check-before-reacting-to-a-lost-one/598166/?ref=bielawski.pro)
- Search Engine Land, 2026-11-19: [AI search meta-analysis: Q4 2026](https://searchengineland.com/ai-search-meta-analysis-q4-2026-502232?ref=bielawski.pro)

## More conversion value does not mean cheaper - watch CPA, not conversions

*Signal: 4 independent sources this week, 8 in 30 days; tracked since 2025-10-01.*

An independent SMEC analysis shows that switching on automated bidding lifts median conversion value by 13%. The cost of a conversion rises by 16%, though, and ROAS, the return on ad spend, stays flat.

That does not mean automation fails, it means you are measuring the wrong thing. Google reports its own numbers: moving from Enhanced CPC to Maximize Conversions with a target CPA delivered 18% more conversions at a similar cost per action. The difference is the benchmark, not the result.

What to do: judge automation by CPA and conversion value rather than conversion count, and prove it in an experiment before you move the whole account.

**Sources:**

- Search Engine Journal, 2026-11-22: [Google Ads AI Max Is Making More Decisions For You - 6 Settings To Audit Now](https://www.searchenginejournal.com/google-ads-ai-max-is-making-more-decisions-for-you-heres-what-to-audit-now/596972/?ref=bielawski.pro)
- Search Engine Land, 2026-11-21: [Google Ads automation drift: Why better performance numbers don't always mean better results](https://searchengineland.com/google-ads-automation-drift-why-better-performance-numbers-dont-always-mean-better-results-502707?ref=bielawski.pro)
- Search Engine Journal, 2026-11-19: [Microsoft Rolls Out Its Biggest Bidding Upgrade In A Decade](https://www.searchenginejournal.com/microsoft-rolls-out-its-biggest-bidding-upgrade-in-a-decade/597963/?ref=bielawski.pro)

## OpenAI shipped two models in one day - one costs four times more

*Signal: 3 independent sources this week, 4 in 30 days; tracked since 2026-09-05.*

On 17 November GPT-6.2 Astra and GPT-6.2 Sol arrived in the API. Astra is a Fable-class model, the top tier of OpenAI's publicly available models, and it costs four times more than Sol.

Astra is priced at $8 per million input tokens and $40 per million output tokens. On bulk work such as product descriptions or summaries, the difference becomes a real cost.

Astra does have one trait that helps in content work: it likes lists, tables and Markdown. Text stored that way is easier to scan for people and for the models that cite it later.

What to do: test Astra on briefs and copy that AI will read, and leave repetitive bulk work to Sol.

**Sources:**

- Jack Roberts, 2026-11-20: [OpenAI Just Dropped a Dumb Genius (Astra 6.2 vs Fable 5.2)](https://www.youtube.com/watch?v=XG%5FYFyGLsXE&t=388s&ref=bielawski.pro)
- Search Engine Journal, 2026-11-19: [OpenAI Publishes Prompting Guidance For GPT-6.2 Astra](https://www.searchenginejournal.com/openai-gpt-6-2-astra-prompting-guidance/597877/?ref=bielawski.pro)
- Bijan Bowen, 2026-11-18: [GPT-6.2 Astra Is INSANE – Does THIS Beat Claude Fable?](https://www.youtube.com/watch?v=jm%5Fh5nRZoCg&t=22s&ref=bielawski.pro)

## Exact match is no longer exact - one in four impressions came from AI Max

*Signal: 3 independent sources this week, 3 in 30 days; tracked since 2026-07-21.*

In an analysis of 383 million impressions, more than 25% of traffic attributed to exact match came through AI Max search term matching (Search Engine Journal).

The word exact stopped meaning the same query long ago, and AI Max pushes it further. Google says the full AI Max feature set delivers 7% more conversions or value at a similar CPA.

Deadlines are moving too. Campaigns with automatically created assets and campaign-level broad match moved to AI Max from September 2026, and Dynamic Search Ads follow in February 2027.

What to do: review your search terms report and add URL exclusions before AI Max sends traffic to pages you do not want to show.

**Sources:**

- Search Engine Journal, 2026-11-22: [Google Ads AI Max Is Making More Decisions For You - 6 Settings To Audit Now](https://www.searchenginejournal.com/google-ads-ai-max-is-making-more-decisions-for-you-heres-what-to-audit-now/596972/?ref=bielawski.pro)
- Search Engine Land, 2026-11-21: [Google Ads automation drift: Why better performance numbers don't always mean better results](https://searchengineland.com/google-ads-automation-drift-why-better-performance-numbers-dont-always-mean-better-results-502707?ref=bielawski.pro)
- Search Engine Land, 2026-11-21: [Microsoft Advertising brings ads to AI Overviews on Bing](https://searchengineland.com/microsoft-advertising-brings-ads-to-ai-overviews-on-bing-502673?ref=bielawski.pro)

## Brand mentions, not links, decide AI visibility

*Signal: 2 independent sources this week, 4 in 30 days; tracked since 2026-11-07.*

In a study of 10,000 brands, every mention type measured was tied to brand presence in ChatGPT, AI Mode and AI Overviews - Reddit posts, YouTube videos and press coverage (Ahrefs).

In a separate analysis of 75,000 brands, mentions had the strongest link to AI Overviews visibility. Brands in the top quartile showed up there more than 10 times as often as the rest (PromptEye).

Correlation is not causation, but the direction is clear: models repeat what many places say about a brand.

What to do: treat Reddit answers, a YouTube channel and industry press citations as part of visibility work, not an add-on to link building.

**Sources:**

- PromptEye, 2026-11-21: [Widoczność w AI — Przewodnik PromptEye v3.1](https://raporty.prompteye.com/api/e-commerce/2026/guideline?ref=bielawski.pro)
- Ahrefs Blog, 2026-11-19: [Barnacle SEO: Spam Risk or Smart Marketing?](https://ahrefs.com/blog/barnacle-seo/?ref=bielawski.pro)

## You can block AI training and still appear in ChatGPT answers

*Signal: 2 independent sources this week, 2 in 30 days; tracked since 2026-10-22.*

Training bots and search bots are separate settings. You can block GPTBot and still let OAI-SearchBot in, the bot that puts your pages into ChatGPT search (Search Engine Journal).

The other way round does not work. Pages that block OAI-SearchBot do not appear in ChatGPT search answers, even when the content is good (PromptEye).

Blocking training sounds safe, but the bot list is longer than it looks. Next to GPTBot sit separate search bots from different vendors: OAI-SearchBot, PerplexityBot, Claude-SearchBot and Bingbot.

What to do: go through robots.txt line by line and decide separately for training, search and user-requested visits (ChatGPT-User).

**Sources:**

- PromptEye, 2026-11-21: [Widoczność w AI — Przewodnik PromptEye v3.1](https://raporty.prompteye.com/api/e-commerce/2026/guideline?ref=bielawski.pro)
- Search Engine Journal, 2026-11-19: [Robots.txt Guide – Best Practices For Search & AI Crawlers](https://www.searchenginejournal.com/technical-seo/robots-txt-guide-best-practices-for-search-ai-crawlers/?ref=bielawski.pro)

## Microsoft rewrote its automated bidding models for the first time in a decade

*Signal: 2 independent sources this week, 2 in 30 days; tracked since 2026-11-19.*

Microsoft Advertising is rolling out rebuilt automated bidding models. The company calls it its biggest bidding update in a decade (Search Engine Land).

The new models are meant to optimize faster and deliver more consistent results. That is a vendor promise, not a measurement from your account, so historical data may stop being a good baseline.

What to do: save results from recent weeks and compare them a month after rollout, before you change bids and targets.

**Sources:**

- Search Engine Journal, 2026-11-19: [Microsoft Rolls Out Its Biggest Bidding Upgrade In A Decade](https://www.searchenginejournal.com/microsoft-rolls-out-its-biggest-bidding-upgrade-in-a-decade/597963/?ref=bielawski.pro)
- Search Engine Land, 2026-11-19: [Microsoft Advertising upgrades automated bidding for faster, more consistent results](https://searchengineland.com/microsoft-advertising-upgrades-automated-bidding-for-faster-more-consistent-results-502328?ref=bielawski.pro)