Press review

Press review 13.11.2026: a new default model in AI Mode, an invalid traffic card, and the maths of funnel economics

Google filtered out ad spend you never knew about, and 72% of US shoppers ask AI for buying advice - though only 17% complete a purchase inside a chat.

Bartosz Bielawski

Bartosz Bielawski

· 9 min read
Share
Press review 13.11.2026: a new default model in AI Mode, an invalid traffic card, and the maths of funnel economics

Topics that clearly more independent sources have covered in the last two weeks than before:

AI Mode answers with a new model - compare data before and after 12 November

Google has made Gemini 4 Hydrogen the default model in AI Mode for users worldwide. The same model now also handles the hardest questions in AI Overviews.

Hydrogen handles multi-part questions, typos and very specific queries better. It also improves query fan-out, so Google shows a wider range of links, including sites the user has never visited.

The change landed two days into the November core update. Google has not said which AI Overviews questions count as "hardest", so you cannot tell how much of the shift in your data comes from the model and how much from the update.

What to do: log two separate dates in your analytics - 10 November (core update) and 12 November (model change). Compare citations and AI Mode visits across those two windows instead of treating the whole month as a single change.

Sources:

The 9-11 November impressions drop in Search Console is a logging bug, not lost visibility

Google has confirmed a logging error in the Generative AI performance report in Search Console. Impressions for 9-11 November 2026 are understated.

The issue affects data logging only. Your actual performance in AI Mode and AI Overviews did not change, so the real number of impressions may have been higher than the report shows. Google has already flagged the problem in the report itself.

What to do: do not blame site changes for a drop over those three days. Mark the period in client reports and exclude it when you calculate your baseline, because it overlaps with the core update rollout.

Sources:

Traffic to the world's biggest sites is down 12.2% since 2022 - that is not a benchmark for your niche

Similarweb, commissioned by SparkToro, counted visits across the top 1 million websites. In the third quarter of 2026, visits were 12.2% below the early-2022 peak.

The peak came in the first quarter of 2022, at 2.243 trillion visits. The low point was the first quarter of 2026 (1.914 trillion), and the third quarter recovered to 1.970 trillion. The share held by the top hundred sites grew only from 33% to 36%.

The data has gaps: the panel tracks browsers, not apps, and the internet-user series was revised after corrections for India and China. SparkToro amended the article after publication because the first version had a wrong chart and a 24% headline figure.

What to do: use 12.2% as global context in conversations about traffic drops, not as a target for a client. Compare your decline with your own figures from last year, not with the world.

Sources:

Notion, Canva and Google Docs are now covered by the spam guidelines - check what you index

Google has expanded its user-generated spam guidelines to cover platforms for building and sharing pages: Notion, WordPress, Canva, Google Docs and app generators including ChatGPT and Claude.

Google recommends blocking indexing of pages that exist only for collaboration or link sharing. It names three tools: the noindex directive, the X-Robots-Tag header and robots.txt.

The spam signals are a sudden spike in published URLs and a small group of users generating most of the URLs.

What to do: review client sites that offer public document sharing or content builders. Block what should not be indexed before Google does it for you.

Sources:

PageSpeed Insights has moved to Lighthouse 13.6 - your score can improve with no code changes

PageSpeed Insights now runs Lighthouse 13.6.0, released on 27 October. The new version simulates page loading differently and estimates FCP and LCP differently.

A page can score better even though nobody changed anything. JavaScript-heavy sites, including single-page apps, will feel it most.

Lighthouse has also fixed the Agentic Resource Discovery audit, which now looks for the ard.json manifest in new locations.

What to do: do not compare pre-October scores with today's in the same report. Judge real performance with field data from CrUX, not the lab test alone.

Sources:

Google Ads has rolled out an Invalid Traffic Overview card globally in Campaigns → Overview. It shows three things: estimated savings, credited cost, and daily and campaign-level reports.

Estimated savings is spend Google never billed because it filtered out invalid activity before billing. Credits issued after the billing period closes, however, do not change cost or CPA in the main campaign table.

A separate Invalid Activity Credit: Search & PMax report in Report editor has 10 columns, including Adjusted cost and Adjusted cost / conv. Credits are attributed to the dates the ads ran, not to the date they were issued.

What to do: open the card and check whether the numbers simply scale with budget - higher savings in a big campaign do not prove the traffic is worse. Verify lead quality in your CRM, because the card only shows what Google detected.

Sources:

A new metric shows how many conversions your first-party data captures

Google has documented the Data Strength Uplift metric. It estimates how many extra conversions your first-party data setup captures, measured over a rolling 14-day window.

Values range from +1% to more than 50%. Google does not simply add up the contributions of individual features; it removes the overlap between them.

You will find it under Goals → Conversions → Summary. It requires at least 100 web conversions or 20 offline conversions in the previous 14 days, and at least 30 days since the feature was implemented.

What to do: pass at least three identifier signals per conversion event. Treat the uplift as recovered measurement, not proof that your campaigns produced more sales - billing and Smart Bidding do not change.

Sources:

ChatGPT Ads adds Target CPA and Target CPC - the old Max Bid settings disappear on 1 January 2027

OpenAI has added Target CPA and Target CPC bidding to ChatGPT Ads. In both modes, the system adjusts bids and delivery to meet your goal.

Product feed campaigns can now target specific regions within a country. Product filters let you combine several conditions on the same field in one ad group.

Maximize Results (Autobid) is being renamed Intelligent Bidding, with the mechanics unchanged. Max Bid and Max CPC will be retired on 1 January 2027, and OpenAI has not yet explained how existing campaigns will be migrated.

What to do: record your own CPA and CPC baseline before the change, because the announcement contains no benchmarks at all. Add campaigns still using the old bid settings to a migration list before January.

Sources:

New Search campaigns in Google Ads Editor 2.14 start with AI Max already on

In Google Ads Editor 2.14, AI Max, text asset automation and Final URL expansion are on by default for newly created Search campaigns.

The release notes do not say that AI Max will be switched on in existing campaigns. Account-level New Customer Acquisition and retention settings have also arrived, including "Only bid for lapsed customers".

The Editor now supports channel prioritisation and budget caps, and Local Inventory Ads in Shopping campaigns are permanently on.

What to do: when you build campaigns in bulk, review the settings before publishing. Automated query matching, asset generation and landing page selection now switch on by themselves.

Sources:

An AI-generated public figure in an ad is banned - a disclosure label will not save you

Google has clarified that advertisers may not use AI to impersonate public figures in ads. Disclosing that the content was generated or edited with AI does not make the ad acceptable.

This is a clarification of existing rules, not a new enforcement policy. The ban targets impersonation, not the use of AI in advertising as such.

The policy falls under the unacceptable business practices category, where consequences can affect the whole account. Google has not announced any new penalties or detection methods.

What to do: add an impersonation check for AI creative to your pre-launch checklist. Do not assume an AI label replaces consent or a licence.

Sources:

AI, GEO and e-commerce

AI citations are not a strategy - tie every visibility metric to a business outcome

SEO practitioners have a new set of metrics to obsess over: AI citations, brand mentions, prompt rankings and share of voice in ChatGPT and Gemini.

Enthusiasm for measuring AI visibility is outpacing understanding of its value. The trouble starts when improving visibility becomes the whole strategy rather than one part of it.

The article's example: adding an FAQ section to every page because "AI likes that format" is a decision made without checking whether the questions matter to readers. The same thing happened in the keyword era.

What to do: attach a business layer to every visibility metric - who you actually reach, whether you influence the decision, and how much of that turns into sales. If you spend more time on how AI should recommend your content than on whether the content deserves to be recommended, change the balance.

Sources:

Checkout inside AI Mode cuts out the affiliate link - the click is your only proof of the sale

Google's Universal Commerce Protocol lets retailers run checkout directly inside AI Mode in Search and in the Gemini app, without a visit to the store's site.

In that flow, the affiliate link disappears. Credit goes only to programmes that can attribute the sale another way: IP hashing, IP matching, assigned discount codes or server-to-server tracking.

Consumers still do not trust AI enough to buy blind, and AI Overviews cannot demonstrate a product. That leaves room for reviews and human-made content.

What to do: keep affiliate links and add a disclosure that a click supports the business. Set up a monthly reconciliation and manually add any conversions the network never reported.

Sources:

72% of shoppers ask AI about purchases, but only 17% complete the purchase in a chat

Optimove's 2026 study: 72% of US consumers use ChatGPT, Claude or Gemini for shopping ideas and gift recommendations, and nearly half do so regularly.

In a McKinsey study of 21,000 consumers across 11 markets, more than half use AI for shopping. In the US, the share of purchases completed inside an AI tool rose from 10% to 17% in a year, a 70% jump.

The agent never sees your ad and does not remember your brand. Each new purchase starts from scratch, so the relationship shifts to the moment after the transaction: order confirmation, first recommendation, return visit.

What to do: give agents your catalogue - products, prices, stock. Keep customer data, order history and loyalty in your own hands, because those decide whether a customer names your brand next time.

Sources:

Marketing, sales and business

A 90 USD acquisition cost against 110 USD of customer value - better creative will not fix that

In this subscription example, a membership costs 22 USD a month and the acquisition cost from ads is 90 USD. At 20% churn, customer value is around 110 USD, leaving 20 USD.

Breaking even would require a fourfold rise in conversion. A realistic goal is 2.5 times, which comes from earning a second transaction within 30 days.

The author lists four conversion points: an onboarding call, direct messages, a weekly Q&A session, and a webinar or sales video. Cheap group ads at 5-10 USD plus upgrading some customers can halve the acquisition cost.

What to do: calculate customer value and churn before you increase the creative budget. If the maths does not work at current conversion rates, the problem is the offer and the path, not the ad headline.

Sources:

White-label SEO at 300-500 USD a month sees less churn than agency services

In a discussion about service models, the claim was that white-label SEO software requires the least effort relative to gross profit. The price point is 300-500 USD a month.

Small-business clients churn less in that model than agency clients do, even though agency services cost more. The reason is simple: less work per client and lower expectations on both sides.

The author's conclusion: choose products with low churn, high revenue retention and high gross margin, then scale the front end.

What to do: compare the margin and churn of both models using your own numbers before adding another agency service. A high price does not make up for a client who leaves after three months.

Sources:

Read more