Home › Briefing › Week of November 7, 2026
AI Search Briefing: Week of November 7, 2026
By Chris Abraham, Gerris AI. Published Saturday, November 7, 2026.
This edition covers the week to Saturday, November 7, 2026: a Turkish investigation into Google AI Overviews and AI Mode, Google folding AI into school Search, OpenAI merging ChatGPT chat and agent work, two Perplexity releases, and two pieces of field research on AI crawlers. Figures a platform reports about itself are labelled as that platform’s own numbers; nothing here runs without a source.
Can a publisher keep Google AI from using its content without losing search visibility?
Turkey opened a formal investigation into Google AI Overviews and AI Mode on Monday, November 2, and ordered interim terms for news publishers while the case runs. The Turkish Competition Authority (Rekabet Kurumu) says it will examine whether Google abused a dominant position with the two AI features. Its interim measure, decided October 8, requires Google to offer news publishers non-discriminatory terms that reflect the economic value of their content in AI Overviews and AI Mode, and to let each publisher decide how its content is used there without any effect on its visibility or indexing in other Google services. The measure covers news publishers, sets no amount and no deadline, and a parallel European Commission investigation has been open since December 2025.
Why it matters to your business: the argument over whether opting out of AI answers can cost you ordinary search visibility is now a legal case in more than one jurisdiction, not a settings page. If regulators establish that the two decisions are separate, every business gets to make its AI decision on its merits. Watch what the United States does with the same question.
Source: Rekabet Kurumu announcement, November 2, 2026, as reported by Maccus
Is there still a version of Google Search without AI answers?
Google made Search and AI one product for schools this week, and confirmed there is no AI-free version of Search to switch to. In its November 3 post on responsible AI for schools, Google says access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be entirely off for K-12 student accounts until a school administrator explicitly turns them on, and that its Guided Learning tutoring is built into Search itself, piloting in select U.S. schools later this month and live in Google Classroom worldwide from November 3. Reporting on the announcement adds the structural fact: Google offers no version of Search without AI Overviews or AI Mode, and its education chief told the New York Times that Search and AI are deeply integrated.
Why it matters to your business: any plan that assumes your buyers use a “regular Google” with ten blue links is planning for a product Google no longer ships. The answer layer is the search product now, in the most cautious rollout Google runs, its schools channel. Being named inside the AI answer is the placement to win.
Source: Google, “Responsible AI for schools with Guided Learning,” November 3, 2026
Is ChatGPT becoming an agent that acts, not just an answer box?
OpenAI is merging ChatGPT’s Chat and Work tabs into a single experience. TechCrunch reported November 2 on OpenAI’s email to users: the merged product rolls out first to a small group for feedback, then widely. In one place, a user can bring in files and plugins and have ChatGPT build a website, make a presentation, or act on the web. Codex stays a separate product, and ChatGPT and Codex will draw their usage from the same allowance pool. Anthropic made the same merge for Claude in September.
Why it matters to your business: the assistant your buyers ask is becoming the assistant that books, buys, and builds on their behalf. The next visibility question after “does it name us” is “can its agent complete the job on our site”: clear prices, clear service pages, and forms an agent can finish.
Source: TechCrunch, November 2, 2026
Do automated systems check the live web before they judge your business?
Perplexity’s Decisions API now searches the web before it answers. Announced November 4, a single setting, mode: "web_grounding", makes the API run a live search for each question and answer from the results. In Perplexity’s own test of 328 true-or-false claims, accuracy rose from 56.4% without grounding, close to guessing, to 97.3% with it; on the public PWS-Decisions benchmark it rose from 51% to 86%, with median response time rising from 139 to 366 milliseconds. Each completed search costs $0.001. These are Perplexity’s figures from its own post; no independent rerun has been published.
Why it matters to your business: software that qualifies accounts, checks eligibility, and verifies claims about companies is starting to check the live web first. When it checks yours, the page it finds is your record. A current page with your services, prices, and locations in plain text is what gets read back as fact.
Source: Perplexity, “Web grounding for the Decisions API,” November 4, 2026
What does cheaper agent research mean for your website?
Perplexity cut the cost of its Agent API research at every effort level and added an ultra level for the widest research jobs. Announced November 5: the medium level’s score on the BrowseComp and DSQA research benchmarks rose from 55% to 77% while its cost fell almost 90%; the xhigh level held its WANDR research score at 76.5% at about a quarter of its August 2026 cost; and the new ultra level runs three research agents in parallel with a fourth merging their findings, scoring 85.9% on WANDR at $21.71 per task. All figures are Perplexity’s own. One mechanism matters for site owners: instead of handing an agent your whole page, Perplexity now extracts only the passages relevant to its task and quotes them exactly.
Why it matters to your business: deep research over the open web, including over your site and your competitors’, just got much cheaper, so expect more of it. Answers assembled by agents are built from quoted passages. Pages with quotable, plainly stated facts about what you sell, where, and for how much are the pages that get quoted.
Source: Perplexity, “Improvements to every Agent API effort level,” November 5, 2026
Why is ChatGPT’s crawler visiting your site less while its referrals rise?
ChatGPT opens small business pages 62% less often than it did in July, and sends those sites more visitors than ever. LovedByAI measured 714 small business websites through its WordPress plugin. Page reads by ChatGPT-User, the bot that opens a page while answering, fell from 8,716 a day in late July to 3,344 a day in late October, a 62% drop overall and 66% at the typical site, in three unannounced steps on August 20, September 17–18, and October 28. Over the same span, visitors arriving from ChatGPT links rose 160% in total and 72% at the typical site. The study’s reading, not confirmed by OpenAI, is that ChatGPT answers more questions from what it has already indexed and cached, and fetches live pages less.
Why it matters to your business: judge ChatGPT by the visitors it sends, never by its bot traffic; a quieter crawler in your logs is not lost ground. The stored copy of your site increasingly does the talking at answer time, which makes crawler access and plain-text facts on the page worth more, not less.
Source: LovedByAI Lab, “ChatGPT visits websites less, sends more traffic,” updated November 3, 2026
Should your robots.txt treat training bots and search bots the same?
No, and 9.2% of large sites now write the file that way: training blocked, AI search allowed. Sitecheckly’s November 2026 survey of the Tranco top 10,000 domains, 4,546 of them with a readable robots.txt, found 9.2% block training crawlers while still allowing AI search crawlers. The distinction is real and named: OpenAI’s GPTBot trains models while OAI-SearchBot feeds ChatGPT search; Anthropic’s ClaudeBot trains while Claude-SearchBot searches; Google-Extended is a training control token, not a crawler, and blocking it has no effect on Google Search or AI Overviews.
Why it matters to your business: a blanket “block AI” robots.txt, often pasted in from a template, can lock your business out of AI answers while doing nothing you intended. Write the file as three separate decisions: training, search, and fetch-on-request, and check the published file, not the draft in your editor.
Source: Sitecheckly, “AI Crawlers List: Training vs. Search User Agents,” November 3, 2026
What is Gerris watching next week?
Three threads. Whether Google answers the Turkish interim measure in public, and whether any U.S. proceeding takes up the same opt-out question. Whether ChatGPT-User page reads step down again, which would confirm the index-first pattern the LovedByAI data shows. And whether the merged ChatGPT chat-and-work experience reaches general users, which moves agent readiness from a developer topic to a storefront topic. The next edition runs Saturday, November 14, 2026.