Signal Observatory

· 18 min

Google pulls Search into AI Mode as Bing puts ads in AI answers

A sidebar of AI searches on Google's homepage, hotel bookings and Lens leading into AI Mode, ads in Bing's AI Overviews almost everywhere, and the core update still rolling on day ten.

Topic radarSelect a topic to filter the archive. The radar shows how this issue's words divide across the topics.

Signal readout

Topic mix

  • AI: 40%
  • Google & Search: 32%
  • PPC: 11%
  • Analytics: 5%
  • Brands: 12%
  • AI40%
  • Google & Search32%
  • PPC11%
  • Analytics5%
  • Brands12%

Share of the issue's words about each topic, counted section by section.

Why it matters

Do not judge the update yet. Two November updates in a row ran well past their two weeks, and Search Console's AI impressions for 9–11 November are lowered by a logging error Google confirmed. Mark the end date when it appears, then compare the same pages over the same number of days before and after.

The short answer

What
Google's AI Mode sidebar, hotel "Continue on Google" and Lens-to-AI-Mode; the core update without an end date; Gemini 4 Neon; Bing AI Overview ads in all markets but China; DEJAN, Seer, Semrush, Verve and DoubleVerify on AI answers and shoppers; Perplexity's Context API; Microsoft's new bidding models; SPUR and LVMH's deal with OpenAI.
Why
Because the entrances to Google Search are being moved into AI Mode one by one, the second search engine made ads beside AI answers the default, and the studies of the week agree that what the web says about a brand still decides whether an AI answer names it.
Who
Site owners and SEOs still reading the November core update; local businesses and hotels; anyone measuring brand visibility in AI answers; retailers; advertisers on Google, Microsoft and Amazon; publishers deciding about AI licences.
Where
Google Search, AI Mode, AI Overviews, Google Images, Gemini, ChatGPT, Perplexity, Bing, Microsoft Advertising, Google Ads and Analytics, Amazon Ads; in the US, the UK, France and worldwide.
When
15–20 November 2026. The core update was still rolling out when this issue was written.
How we know
Google's Search Status Dashboard and blog; Search Engine Roundtable, Search Engine Journal and PPC Land for sightings, emails and write-ups; the studies' own pages from DEJAN, Seer, Semrush and Ahrefs; Microsoft Advertising and Perplexity's own posts; Press Gazette and TechCrunch.

In plain words

Google is turning more and more of its search page into a chat with its AI. This week it added a list of your recent AI chats on the homepage, and buttons that move you from hotel results and from pictures into that chat. Bing, Microsoft's search engine, now shows ads next to its AI answers in almost every country. Google's big ranking shake-up, which started on 10 November, was still going on. And new studies looked at how AI chatbots choose which shops and brands to name: mostly from what they find on the web at that moment, and the biggest shop is not always the one they mention.

Google spent the week moving doors. On 19 November a sidebar of recent AI searches went global on the desktop homepage, hotel results started sending people into a booking flow inside AI Mode, and Google Images got a button that turns a picture into an AI Mode chat. The same day Microsoft said ads now run in Bing's AI Overviews in every market except China, with no new campaign to set up. Underneath both, the November core update was still rolling out on day ten. And the week's studies, from DEJAN, Seer and Verve, asked the question every brand now asks: when an AI answer names a business, where did it get the name from?

The core update, day ten

Google's Search Status Dashboard still showed only the launch entry on 20 November: "Released the November 2026 core update, which applies globally and to all languages. The rollout may take up to 2 weeks to complete." It started on 10 November at 10:07 PST; every other 2026 ranking incident on the summary page has a finished duration, this one does not.

Semrush's explainer of 17 November adds no winners and losers — "Google hasn't shared what changed" — but puts the calendar in perspective: two weeks would run "until November 24", while the November 2024 update took "nearly 24 days" and November 2023 "almost 26". No new domain-level analysis was published between 15 and 20 November.

How long Google's core updates took to roll out, days

November 202326.0
November 202424.0
December 202518.1
March 202612.2
May 202611.9
2025–2026 from Google's Search Status Dashboard (18 days 2 hours, 12 days 4 hours, 11 days 21 hours); November 2023 and 2024 as Semrush rounds them ("almost 26", "nearly 24"). The November 2026 update started on 10 November and had no end date on 20 November.

More of Google Search leads into AI Mode

Four changes in one day, 19 November, all reported by Search Engine Roundtable:

  • A sidebar of AI searches. Robby Stein, Google's vice president of product for Search, on X: "we're rolling out a collapsible sidebar to easily find & jump back into your recent AI searches." His reason: "People are coming to Search with more questions that aren't just one and done." It is now global on the desktop homepage; Barry Schwartz first saw it on 30 October.
  • "Continue on Google" on hotels. A button on hotel listings opens AI Mode's booking flow. In Lluc B. Penycate's screenshots Expedia's rate through that flow, $170, was promoted over the hotel's own direct rate of $158: "The cheaper direct rate is the one you can't see." Schwartz could not reproduce it himself, and wrote that "It is not good that you don't get to see all the rates."
  • Lens to AI Mode in Images. In the Google app's Images tab an AI Mode + Lens icon opens an AI Mode chat about the picture. Outside the app Schwartz was "taken to Google AI Overviews" instead, and he sums up the direction: Google keeps "funneling everything into AI Mode."
  • AI emojis in "What people are saying". The emojis beside those results come from Google's AI descriptions, not from the sites cited. Schwartz reproduced it.

Google's own holiday post of 16 November, by Craig Ewer, explains the bookings: "AI Mode in Search can map out your trip and take care of reservations." Via PPC Land, hotel booking is US-only and English-only, with ten partners — Booking.com, Choice, Expedia, Hilton, Hotels.com, IHG, Marriott, Priceline, Trip.com and Wyndham — and Google's help page says "Google isn't a party to the booking transaction and doesn't send confirmation emails."

On 18 November Google DeepMind introduced Gemini 4 Neon, between Carbon for the hardest long tasks and Hydrogen for fast everyday ones, with a one-million-token context window. It is in the Gemini API, AI Studio, Antigravity and the Gemini app for AI Pro and Ultra. The post does not mention Search: Hydrogen is still the model behind AI Mode.

Dan Petrovic of DEJAN followed up his Hydrogen work the same day. He ran 10,701 pairs of prompts, 3,503 distinct ones about 119 sites, between 13 and 18 November, each once from the model's memory and once with Google Search grounding. "Search supplied 45% of the brands in each answer": 3.4 of 7.5 on average. "Only 1.8% of answer pairs are identical", and a brand that stayed kept its position only 39% of the time. Out-of-date brands rarely survive the search: "of 70 out-of-date brands named in those passages, 1 reached the final answer." ESPN BET was named 226 times from memory and never with search. In a companion post Hydrogen, from memory, named none of DEJAN's own products across ten lists.

Seer Interactive looked at ChatGPT from the inside on 18 November. Nick Haigler read the reasoning of GPT-6 Luna on 20 prompts through the API, which he calls "directional". Fewer than one brand in ten was named before the first search; 59% appeared right after the first round of broad searches. site: queries made up 13% of the first round and 75% of later rounds: the model checks brands on their own sites before it decides. It shortlisted about eight options and mentioned about six; 38% of its final notes gave a reason to drop one, and nine of the ten brands it flagged for trimming did not appear in the answer.

StudyWhat was askedWhat decided the brands named
DEJAN, Gemini 4 Hydrogen10,701 prompt pairs, memory vs Search groundingSearch supplied 45% of brands per answer; 1.8% of pairs identical
Seer, ChatGPT (GPT-6 Luna)20 prompts, reasoning readUnder 10% named before searching; 75% of later searches were site:
Semrush, local AI Overviews480 searches, 12 US markets373 of 1,023 businesses recommended were not on page one
From DEJAN and Seer Interactive (18 November 2026) and the Semrush blog (17 November 2026). Different models, prompts and sample sizes: read each row on its own.

AI Overviews take the place of the map pack

Chris Shirlow, an agency owner writing on the Semrush blog on 17 November, ran 480 searches by hand for ten local services in twelve US markets between September and November. "In my first round, 38% of searches showed an AIO. By my last round, 99% (79 of 80) showed them." And: "On 79 of 80 searches, there was no separate map pack at all." Phrasing mattered: "conversational searches got an AIO 66% of the time, compared with 24% for bare ones." The AI Overviews did not simply copy the local pack: "More than a third of the businesses the AIO recommended (373 of 1,023) didn't appear on page one of Google." The sample is small and the last round was captured on a single day, which the author himself says could be a test.

Try it on your own page

See what Vectorscope finds on a page of yours

Everything above is measurable on your own work, and the check takes about a minute. Paste one address and read what comes back.

Open Vectorscope

Free · no account · one at a time

ChatGPT: a new model, and business agents

OpenAI released GPT-6.2 Astra and Sol in its API on 17 November. Search Engine Journal's write-up of the prompting guide notes that Astra is "more concise and direct by default than earlier models"; nothing was said about ChatGPT's search or citations, so there is no change to report there yet.

Search Engine Roundtable took a first look at ChatGPT's business agents, the follow-up to last week's Business Hub. The builder says "We're gradually rolling out this feature to different types of businesses." Wayfair's agent showed product cards from a catalogue feed; Chewy's answered "I don't have a live Chewy product catalog connected." Users see "Business can't see your chats or your personal info". Lily Ray's verdict: "it pulls a lot of recommendations out of your website so it's pretty low-effort to get it set up." Which is Seer's finding from the other side: what your own site says is the input.

Shoppers and AI answers, measured

Verve compared US retailers' share of gift searches with their share of AI shopping answers between 1 September and 16 November (via PPC Land, 19 November). Amazon took 24% of the searches and 11% of the answers; Target, Best Buy, eBay, Gap and Etsy all did better in AI answers than in search. AI gift prompts had a median of 24 words against 9 for searches, and named a budget in 7.2% of cases against 0.49%. Verve's own footnote: "a mention is not a recommendation". The AI platforms and the number of prompts are not given.

RetailerShare of searchesShare of AI answers
Amazon24%11%
Walmart12%9%
Target5%8%
Best Buy2%3%
eBay2%3%
Gap1%2%
Etsy0%2%
Verve, US, 1 September–16 November 2026, via PPC Land (19 November 2026). The AI platforms measured are not named.

How many people ask an AI at all? DoubleVerify's survey of 22,000 consumers in 21 countries (Sapio Research, February 2026; report 18 November) found 17% use AI chat for product research, against 39% for search engines and 40% for online reviews. "AI chat is increasingly part of these conversations, especially for older Gen Z and younger Millennial consumers."

Where consumers research products, % of respondents

Online reviews40
Search engines39
Physical store30
Comparison sites26
Friends and family26
Social media26
Video reviews24
AI chat17
Influencers11
DoubleVerify, 22,000 consumers in 21 countries, fieldwork by Sapio Research in February 2026, margin of error ±0.7 points; several answers allowed. Via PPC Land, 19 November 2026.

Two more measurements, both with caveats. Semrush launched an "AI-influenced traffic" metric on 18 November: a modelled estimate that counts direct and branded-search visits within a "24-hour window" after an AI mention, not only clicks; it is the vendor's own model, US and UK first. And PPC Land quoted a SISTRIX figure on 19 November: across 10,000 US AI Mode keywords Wikipedia was the most cited domain, with 12.02% of appearances, ahead of google.com at 7.41%.

Bing puts ads in AI Overviews almost everywhere

Microsoft's November roundup, 19 November: "Ads in AI Overviews are now available in all markets except China." And "There's no separate campaign set-up required": existing campaigns, Performance Max and AI Max included, are eligible. "Ads in AI Overviews on Bing are the latest step in bringing ads to AI experiences, building on ads in Copilot." The same roundup says Microsoft's new Discover feed shows "35% fewer ads per page" than classic MSN, with higher viewability and click-through by its own October data.

The day before, Microsoft rolled out what it calls its biggest bidding upgrade in a decade, applied to Smart Bidding automatically (Search Engine Journal, 18 November). Its internal numbers are below. Navah Hopkins: "the system is much more responsive now!" — and "Every account is different so a blanket recommendation still needs to be put to the test." Microsoft's guidance is to move targets in steps of 15% and judge on about 30 conversions in 30 days.

StrategyReported uplift
Maximize Conversion Value or Target ROAS+10% conversion value per click
Target CPA moved to Target ROAS+8% conversion value
Enhanced CPC moved to Maximize Conversions with a CPA target+18% conversions
Microsoft Advertising's internal data, via Search Engine Journal, 18 November 2026. Not independently measured.

PPC elsewhere

  • A new ad at the bottom of AI Mode. PPC News Feed, crediting Sachin Patel, 19 November: "Google is testing a new ad format positioned at the bottom of AI Mode", with more content than the earlier text-link tests.
  • Hotel campaigns move into Search. Hotel campaigns migrate to Search campaigns for Travel from Q4 2026, with new AI Max for Search help pages for travel; Performance Max setup now shows Channel Prioritization and Promo Mode (Search Engine Roundtable, 18 November).
  • Budgets by platform in Analytics. "Cross-channel budgeting in Google Analytics now supports source platform dimensions", 19 November; the help page says it "won't affect budgets or spend in your connected accounts."
  • AdSense for Search. A new publisher policy from 17 January 2027: these ads only on results pages, one ad request per user action, and never more ads than organic results on a page.
  • Satya Nadella on the agent web. At the DealBook Summit, 18 November, via TechCrunch: "if you were monetizing ads because of the eyeballs, you now have to probably get down to essentially maybe CPC wins", and "if you don't accept the agent, somebody else will."

Publishers and the open web

DMGT, the Daily Mail's owner, joined SPUR, the Standards for Publisher Usage Rights coalition, on 17 November (Press Gazette). SPUR has more than 80 members and a board with the FT, the Guardian, the Telegraph, the BBC, Sky News, Mediahuis, AP and now DMGT; it wants standards to set, verify and enforce how AI uses publishers' content. Owen Meredith of DMGT called scraping a problem "that no news organisation can combat alone".

Going the other way, LVMH's media division signed with OpenAI on 16 November (via PPC Land): 17 titles including Les Echos, Le Parisien and Paris Match, reaching "more than 27 million" people a month, will appear in ChatGPT as summaries with links back to the articles. Price, length and training rights were not disclosed.

Perplexity opened a Context API on 17 November that reads up to ten URLs per call and returns their text, using Perplexity's own copy when it is under two days old and fetching live otherwise. The post says nothing about which user agent fetches or how it treats robots.txt, which is the question a site owner will ask first. No new EU action against Google and no new publisher lawsuit were published this week.

Technical SEO

  • Rank Math: 16 vulnerabilities this year. Search Engine Journal, 18 November: "more than any other SEO plugin", the worst two allowing remote code execution (CVSS 8.8) in 1.0.276 and earlier, plus an unauthenticated stored XSS rated 7.2. If you run it, update.
  • CDNs and crawlers, the quotes. Search Engine Journal's write-up of Google's CDN guidance from last week adds Gary Illyes: "Basically, CDNs take the valve off the hose, so now crawlers are drinking from a fire hose." And Cherry Prommawin: "In panic, one might log into their CDN dashboard and turn on the setting to block bots entirely."
  • Data Feed validation tool retired. "The Data Feed validation tool has been turned down and is no longer available", 19 November; the Book actions docs now have a section on checking feeds for common errors.

No link-building news this week. The one useful piece was Ahrefs' guide to "barnacle SEO" on 18 November: "the strategy of getting your brand featured on web pages that already rank", which is also how brands enter AI answers. It draws the line at parasite SEO: "Google treats this as site reputation abuse". Read beside DEJAN and Seer, the advice is the same from three directions: be mentioned on pages search already trusts, and say the facts on your own site.

Our news: GLLmolfar 0.9.38–0.9.46

Eight versions of GLLmolfar, our desktop crawler, came out between 15 and 19 November. The ones that matter in a core-update week:

  • Traffic diagnosis. 0.9.40 adds a report that finds the days your Search Console clicks changed, splits each change into impressions, position and the rest, marks Google's updates on the chart and forecasts seven weeks. No AI model is used: every number comes from a formula shown beside it. 0.9.41 adds a calendar of search and AI events to the start page, with what is running now.
  • What an AI bot sees. 0.9.45 compares every page before and after JavaScript — text, prices, tables, FAQ, structured data — and checks the key pages after a crawl. New in the AI tabs: contradictions between your own pages, Google × AI, and AI Mode fan-out queries found in Search Console.
  • Better topics and languages. Repeated template text is left out of the embeddings, so topics group pages by what they say; on sites in several languages, 0.9.46 finds topics in the main language and puts each translation in its original's topic, and Similar no longer counts translations as near-duplicates.
  • Big crawls. 0.9.42 needs about a third less memory at 80,000 pages. Stop now ends a crawl at once, failed pages can be re-crawled in one click, and embeddings have their own Stop. In alpha, for signed-in users: write the crawl to Google Sheets as it runs, in a spreadsheet only your account can see.

Also this week

  • Lily Ray on self-ranking lists. Via PPC Land, 19 November: brands that rank themselves first were "left out of the recommendation in 83% of Google AI answers citing their pages", up from 69% in June.
  • Perplexity's research lab. Frontier Research, 16 November, "supported by considerable computational resources, in excess of 10,000 GPUs."
  • Brands rated by a model, without the web. GPT Insights' Brand Atlas USA, 18 November, asked a model 200 questions about each of 12,088 brands in 14 countries with web search off; in the US YouTube ranks first for recommendation and LinkedIn second. These are a model's opinions, not mentions in AI answers.
  • "Show original results". Google is testing the button for typo-like queries; Schwartz could not reproduce it. 19 November.

Questions people ask

Has the November 2026 core update finished?

Not as of 20 November 2026. It started on 10 November and Google said it may take up to two weeks, which ends around 24 November; the November 2023 and 2024 updates took about 26 and 24 days. Search Console's AI impressions for 9–11 November are lowered by a logging error.

What is the AI Mode sidebar on Google's homepage?

A collapsible list of your recent AI Mode searches on the desktop homepage, rolled out globally on 19 November 2026, so you can return to a conversation instead of starting a new search.

Do I need a new campaign for ads in Bing's AI Overviews?

No. Microsoft says existing campaigns, including Performance Max and AI Max, are eligible, and the ads run in all markets except China since November 2026.

Does ranking in search still matter for AI answers?

Yes. DEJAN found that Google Search grounding supplied 45% of the brands in each Gemini 4 Hydrogen answer, and Seer found ChatGPT named fewer than one brand in ten before it searched, then checked brands on their own sites.

Why is my brand mentioned by AI but not recommended?

A mention is not a recommendation, as Verve's study says. In Seer's test ChatGPT shortlisted about eight options, mentioned about six and gave reasons to drop some; Lily Ray found self-ranked lists were left out of 83% of the Google AI answers that cited them.

Sources

Every claim above is dated and linked. Where a number comes from a measurement rather than an announcement, whoever measured it is named.

The product

Watch it on every page, every day

One reading tells you where a page stands today. The product asks the same questions of the same assistants continuously, so a change is something you are told about rather than something you go looking for.

Request an invitation

Invite-only while we keep the readings honest

On this topic

What changed in classic search?

Every day Search Console beside the assistants' answers. How the product works

Author

Olexandrauthor

15+ years in marketing and SEO. Dozens of products, mostly the US market. Research and measurement.

LinkedIn

Pass it on

Read in your language

Opens a browser translation of this English article. The original source stays in English.

All issues