Press review

Press review 07.11.2026: Brave's index goes to AI models, and backlinks stop being the currency of search

Updating a YouTube description was linked to 359% more AI retrievals, and across 75,000 brands, mentions correlated three times more strongly than backlink count. TikTok has just added a full keyword layer.

Bartosz Bielawski

Bartosz Bielawski

· 7 min read
Share
Press review 07.11.2026: Brave's index goes to AI models, and backlinks stop being the currency of search

Brave's index went to AI models, and you cannot switch it off without dropping out of Google

The New York Times sued Brave on 2 November, claiming the company copied more than 30 billion webpages without identifying itself and sold the content to AI companies. Cohere, Mistral, Perplexity and You.com all draw on that index. The same day, the Daily News, the Chicago Tribune and the Orlando Sentinel filed a separate suit.

The most important detail is technical: Brave's crawler does not use its own user agent. You cannot block it in robots.txt without also blocking Googlebot. The only way out is noindex, which also removes the page from Google.

What to do: run site:yourdomain on search.brave.com and add a separate line to your audit for presence in indexes sold to AI models. Blocking named AI bots such as GPTBot or Bytespider still works bot by bot, but it won't cover Brave.

Sources:

In a study of 75,000 brands with DR above 40, brand mentions across the web correlated with AI Overviews presence roughly three times more strongly than raw link count. Volume metrics, meaning total links and referring domains, fell to the bottom of the ranking. Branded anchor text on its own came in as the second-strongest factor.

The benefit is extremely unevenly distributed: brands in the top quartile had 169 AI Overview mentions, while those in the second quartile had 14. About 26% of the brands had no mentions at all, even though all of them had DR above 40. In a separate dataset, the share of top-10 organic URLs among AI Overview citations fell from 76% to 38%.

What to do: replace adjectives in your content with facts in a subject-action-measurable result format. Every paragraph must stand on its own, because the retriever splits a page into passages of about 100 words and pulls out only one.

Sources:

Editing a YouTube description was linked to 359% more AI retrievals

In an experiment on about 60 videos, descriptions were updated and manual timestamps added to 15 of them. Videos with updated descriptions then saw 359% more daily AI retrievals. Videos left unchanged grew 17%, and YouTube as a whole grew 25%.

In Google AI Overviews, the increase was 293% versus 27% for the site as a whole, and in AI Mode it was 129% versus 11%. The authors note that they could not isolate the effect of the description update itself, so this is correlation, not proof of causation.

What to do: fill out the descriptions of your most important videos, using more of the 5,000-character limit, with chapters and the terms people actually search for. You're only changing the description, not the video, so the test is cheap.

Sources:

OpenAI will give businesses a counter for their own ChatGPT mentions, but without prompts or competitors

OpenAI is working on tools in Business Hub that will show a business how it performs in ChatGPT answers. The counter measures two things: how often the business appeared in answers and how many clicks its site links received. Data is aggregated daily and displayed as a chart.

Access is limited to businesses that have verified ownership, for example by connecting a Shopify store. It does not include prompts, user data or competitor benchmarks. The first signs of the feature appeared in August, but the tool is not publicly available yet.

What to do: verify your business ownership in Business Hub now so you are in the first wave. The counter shows the effect, not the cause, so without your own library of questions, you still won't know what to change.

Sources:

Google Maps now shows a chat button next to Call - check that messages to your business actually arrive

Google Maps on Android and iOS now shows a Message button next to Call. That makes chatting with a business one of the two main buttons on a Google business profile. The hours editor has also gained a Generate hours field that turns a plain-language description into filled-in opening hours.

In a 5 November bulletin, Google announced that Gemini tools for business profiles are also coming to Workspace accounts. The exclusion of the EEA and the UK and the 12-language limit are gone. The connection works only for the owner or manager of a single verified profile, and only in the Gemini web app. Google's help page still says you cannot connect a profile from a work account.

What to do: before the holidays, send a test message through the chat on every profile you manage. A broken chat in one of the two main buttons is the easiest way to lose a customer during peak season.

Sources:

A model can refuse to answer questions about your industry, and that is also a visibility variable

In a UK AI Security Institute test, three Anthropic models refused more than half of a set of reasonable AI safety research tasks. The Meta Oversight Board also found that several popular models were more likely to reject queries about repressive governments, so refusals have a political dimension, not just a technical one.

Refusals cost vendors too: one type of safety classifier added 24% to compute costs in Anthropic's chatbots. Google's guidance on AI-generated content requires manual fact-checking before publication.

What to do: keep a monthly refusal log covering a few dozen prompts in your category. A drop in visibility and a change in model behaviour are two different problems that need different responses.

Sources:

TikTok copies Google Ads keywords, but its negatives don't catch plurals

TikTok uses the same match type names as Google: broad, phrase and exact. Phrase and exact match are based on the meaning of the search rather than word order, so "luxury fragrance" will also catch the name of a specific perfume. An ad group can hold up to 1,000 keywords, though TikTok recommends at least 20.

The differences lie in the details that break reports. Negatives do not automatically block plurals, typos or synonyms, so each variation has to be added separately. Automated Keywords is on by default for SMB accounts, and TikTok does not notify you when new keywords are added. Keyword-level bidding works only in Traffic campaigns with a Cost Cap bid.

Search campaign metrics are delayed by about 8 hours, most conversions show up after three days, and SKAN data is not available at all. You also cannot break down the keyword or search term report by creative.

What to do: start with the search terms report and negatives for each variant, then check the Automated Keywords toggle. In cross-platform reports, label the "TikTok search" row to state explicitly what is missing.

Sources:

Reddit has started selling ads in search results, with no keyword mechanism

Since 3 November, Reddit has been testing ads in search results. The ads are meant to appear based on contextual relevance to the query, and the mechanics as described include no keywords or match types.

That is the opposite of TikTok, which is adding a full keyword setup. For planning, it means one thing: today a Reddit budget buys topic intent, not a query.

What to do: if you test Reddit, report results by thread topic rather than by query. Without keywords, you cannot make an honest comparison between the channel and Google Ads.

Sources:

Marketing and content

In 92% of top Instagram search videos, the product is named in the first few seconds

In an analysis of about 22,700 public beauty posts across four categories, 92% or more of the top videos named a specific product within the first few seconds. Most of them were tutorials that answered both "how do I do this" and "which product should I pick".

Follower count showed no clear relationship with appearing at the top. Accounts focused on a single topic appeared again and again, and hair and nail tutorials almost always named the target audience. The study captures directional patterns only, because Instagram personalises results.

What to do: take the next video on your calendar and give it a more specific subject. Name the product in the first second, both in voiceover and on screen, and end with a question only the viewer can answer.

Sources:

Seventh place on page one is not visibility, because on mobile, ads and the map sit ahead of you

Google's first page is not a single spot; it is a ladder. On mobile, local results stack in a fixed order: ads, a map with a few businesses, and only then the regular results. So position seven is effectively well out of sight.

In the case described, a basement finishing company used a new site and optimisation to move its main queries from positions 6-7 to 1-2. The published result is 97 leads worth $485,200.

What to do: run a scroll test in a private window on your phone, searching for the service and the city rather than the business name. Count how many ads and map listings appear above your page.

Sources:

E-commerce

A new category page only makes sense when neither the old page nor the filter answers the buying question

The starting point is Search Console: it shows which queries bring traffic and which pages get clicks. If the same URLs rank for both the broader and the narrower query, it is usually better to cover both with one page than to build a new URL.

The number of items in a category can be misleading. Twenty listings may be mostly colour variants or leftover sizes, and availability has to be checked across several restocking cycles, not on delivery day. Seasonal collections belong in promotions rather than on a separate page that sits empty a month later.

What to do: before creating new URLs, check the overlap between organic results and your on-site search. A new page must offer side-by-side comparisons of specs, prices or use cases that the filtered view does not have.

Sources:

Read more