An expectation is not an agreement
A court ruled that Google never promised publishers any traffic. In the same week Google wrote down what it now counts as quality, and a study found that the median AI citation loses half its weight in eleven days.
The short answer
- what
- The second phase of Google's September 2026 spam update, rewritten guidance on helpful and AI-generated content, the dismissal of Penske Media's AI Overviews suit, studies on AI citation decay and on AI shopping answers, and a run of new ad formats.
- why
- Because the week moved one assumption from habit to record: nobody owes a site traffic, and nobody owes a brand a mention. What remains is what a site does on each page, and how often it is measured.
- who
- Sites that publish at scale or with AI help, publishers who expected search traffic in return for their content, brands that depend on being named in AI answers, and advertisers on Google Ads and ChatGPT.
- where
- Google Search, AI Overviews, AI Mode and Gemini; ChatGPT, Claude, Perplexity and Copilot; Google Ads and ChatGPT Ads; the US District Court for the District of Columbia; the EU General Court.
- when
- 28 September – 4 October 2026. Spam update phase two from 30 September; the court ruling and the guidance changes on 1–2 October; Google's Barcelona remarks reported 4–5 October.
- how
- Google's documentation and Search Status Dashboard; Search Engine Roundtable, Search Engine Journal and PPC Land; Profound's citation study; Vaer AI's retailer study; Graphite's Similarweb-based ChatGPT ads report; Digiday on the SPUR standard.
In plain words
For years websites let Google read their pages, and Google sent people back. Many sites thought of that as a deal. This week a judge said it never was: Google had not promised anyone any visitors. In the same week Google explained in writing what it thinks a good page is, and started removing pages made in bulk. And people who study AI answers found that when an AI names a website, it usually stops naming it about a week and a half later. So being mentioned is not something you earn once. You have to keep earning it.
Two weeks ago Google moved where the click leads. Last week everyone else measured what that cost. This week a judge was asked whether Google owed anything for it, and said no.
The ruling came on 1 October, in a week in which Google also published, for the first time in its site-owner documentation, what it means by the quality of a page. Its spam update hit a second time. And outside Google, the most useful number was not about traffic at all: it was about how long an AI answer keeps naming you.
The spam update comes back for a second pass
The September 2026 spam update, announced on 24 September, is still running. On the Search Status Dashboard it has no end date yet. Google has said this rollout will take about two weeks rather than the usual two days, and the second week brought a second wave. Glenn Gabe saw it land on Wednesday 30 September.
The update continues to roll out and I believe we saw a specific change on 9/30.
Glenn Gabe, via Search Engine Roundtable, 1 October 2026
The first wave, from 25 to 27 September, fell mostly on templated and programmatic sites. Lily Ray's example was AI-generated pages with one page for every phone number or area code. The second wave was harder to read. Reddit's machine-translated pages, which had collapsed in the May and June updates, rose again in Italy, Germany, Brazil and France. Publishers on WebmasterWorld reported that Discover and News traffic had "again vanished". Fourteen trackers showed movement on 30 September, from Mozcast and Semrush to SISTRIX.
The trackers' own caveat still holds. SISTRIX's incident for reduced Google data collection, open since 27 August, has had no resolution. Its last entry, from 26 September, still says Google is returning "incomplete or distorted" results in some cases.
Google writes down what it counts as quality
On 1 and 2 October Google rewrote two pages of its Search documentation. Neither change appeared on its own documentation changelog. Both are more specific than anything it has published for site owners before.
The page on helpful content now defines main content as "any part of the webpage that directly helps the page achieve its purpose". It also takes four attributes from the quality rater guidelines: effort, originality, talent or skill, and accuracy. And it has a new paragraph that, according to Search Engine Journal, had previously appeared only in the rater guidelines, since at least May 2025.
Fabricating creator profiles (such as by using AI-generated headshots, made-up names, or false credentials to make content appear as if it was written by human experts) is a form of deception.
Google Search Central, "Creating helpful, reliable, people-first content", October 2026
The page on using generative AI content changed the same week. It now says it is "critical to manually factcheck and review all AI-generated content", and it explains why: "generative models don't retrieve facts, but predict a likely sequence of words." Google is not banning AI-written pages. It is saying a person has to stand behind each one.
Gary Illyes said the same thing at Search Central Live in Barcelona, as Search Engine Roundtable reported on 5 October. Google filters out 40 billion spam pages a day. It now uses AI to find AI spam. It runs more spam updates because there is so much more new content. And scaled, low-effort content has become a bigger concern than link spam. It also judges quality page by page, not domain by domain, so one site can have pages that rise and pages that fall.
In the same talk Illyes gave rare figures for how long Google's own processes take. Search Engine Journal noted that he gave no sample size or period, so these are indications from Google, not measured guarantees.
| Process | Typical time, per Google |
|---|---|
| Discovering a new URL | About 20 hours |
| Indexing, end to end | About 1.5 hours |
| Updating a title or snippet | 1–2 days |
| Processing a sitemap | About 24 hours, up to 14 days |
| Refreshing a known URL | About 30 days |
| A site move | 1–3 months; at worst 6 months to a year or more |
| Recovering from a core update | 3–6 months; at worst until the next core update |
Try it on your own page
See what PatternScope 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 PatternScopeFree · no account · one at a time
"An expectation is not an agreement"
On 1 October Judge Amit Mehta of the US District Court for the District of Columbia dismissed Penske Media's antitrust suit against Google over AI Overviews. Penske, which publishes Rolling Stone and Variety, argued that Google used its dominance to take publishers' content for AI answers, and that in exchange publishers had been led to expect traffic. The complaint cited a 34.5% drop in click-through rate associated with AI Overviews. The judge's answer was that publishers had "failed to plead any actual agreement whereby Defendants promised to 'sell' Plaintiffs any specific amount of traffic." Then, in six words: an expectation is not an agreement.
He did not deny the harm. The ruling refers to "the knock-on consequences to journalists, educators, and other online creators whose content Google takes and repurposes without compensation." What it denies is that the law, as pleaded here, can do anything about them. PPC Land does not say whether the dismissal is with or without prejudice.
Two other stories this week show what the arrangement looks like when nothing is promised. On 30 September The Information reported, and Search Engine Roundtable relayed, that Google's AI contribution pilot pays about 100 publishers when their content is used in AI Overviews, AI Mode or Gemini answers. One large participant earns more than a million dollars a year. For smaller sites the numbers are very different.
For several small to midsize blogs and websites testing the programs, the payments amount to less than one-tenth of 1% of their advertising revenue.
The Information, via Search Engine Roundtable, 30 September 2026
And on 29 September Google appealed to the EU General Court against the Digital Markets Act decisions requiring it to share search data with rivals by January 2027. The same day Nick Fox, who runs Google's Knowledge and Information unit, said in an interview that "it literally makes me sad that we need to degrade our product in this way." In Europe Google has already rolled back features and its site reputation abuse policy. The result is that a search page in Europe now behaves differently from one in the US. A report that compares markets has to take that into account.
An AI citation has a half-life of eleven days
The most useful study of the week for anyone measuring AI visibility came from Profound on 30 September. It followed 883,000 pages across seven engines for a year — ChatGPT, Perplexity, Claude, Google AI Overviews, AI Mode, Gemini and Copilot. That is 1.19 million page-and-engine lifecycles, from 8 September 2025 to 8 September 2026. The headline is in the title: the median page loses half its citation share within 11 days of its peak.
| Finding | Value |
|---|---|
| Median citation half-life | 11 days |
| Cited pages that fall to half their peak within two weeks | 78% |
| A Claude citation still relevant after 8 weeks, against ChatGPT | 2.8× as likely |
| Pages cited by 5+ engines still above half their peak after 8 weeks, against 1 engine | 5.5× as likely |
| Correlation of citations between engines | 0.03–0.09 |
Authority in AI search expires rather than accumulates.
Profound, "The half-life of an AI citation is 11 days", 30 September 2026
The last row matters most. A correlation between 0.03 and 0.09 means the engines decide almost independently: being cited by ChatGPT tells you nearly nothing about whether Gemini or Perplexity will cite you. Taken with the half-life, it means that a single snapshot of citation share, taken once and on one engine, describes one moment and one engine.
A second study, from Vaer AI for Lightspeed Commerce on 29 September, shows what the engines decide when they do decide. It covers about 500,000 answers from ChatGPT, AI Mode and AI Overviews to shopping questions. Without web search, the AI named a national chain about two-thirds of the time. When shown a large and a smaller retailer with no labels, it picked the larger one 90–94% of the time. About ten chains took 40–44% of all recommendations. With web search on, chains still got about half, and local shops about a quarter. Small retailers appeared in the cited sources far more often than in the recommendations. One word changed the result: asking for independent stores raised small retailers from about a third to nearly four-fifths. Asking for local or near me did not help, because a nearby branch of a chain counts as local.
Brands get an AI Overview, and Gemini starts to label its links
On 30 September Chris Long reported that Google had started showing AI Overviews on brand searches: "93% of the brands I tested have AI Overviews appearing." He tested about a hundred brands, among them Reddit, Salesforce, Amazon and Adobe. Most of the summaries sit partway down the page, below the brand's own result. Google's own name and news publishers were the exceptions. DemandSphere's data agrees: AI Overviews on branded queries "tripled in September".
A search for a brand's name used to be the one result the brand controlled. Now it often comes with a summary of the brand written by Google's model, from sources the brand did not choose.
On attribution there was some good news. On 1 October Search Engine Journal reported that Gemini had started adding UTM parameters to its outbound links. Google has not documented it, and John Mueller confirmed only that he saw it too. It matters because, according to the Reddit thread that found it, the Gemini mobile app passes a referrer only about 9% of the time, and Android assistant calls strip it entirely. That traffic had been landing in Direct. One warning: a similar report the next day, of tracking parameters on AI Overview links, was withdrawn. It turned out to be a Chrome extension.
Advertising moves inside the answer
The ads side of the week had the most launches. Graphite published the first public CTR figures for ChatGPT ads on 29 September, using Similarweb panel data across nine markets. Its definition of CTR is clicks per exchange that carries an ad, meaning one user message and ChatGPT's reply.
ChatGPT ads: click-through rate by market, % (Graphite, Similarweb panel)
The markets where ChatGPT ads launched later have higher click-through rates. The US, where they began, sits below 1%. In the same report B2B SaaS took 19.9% of US ad impressions between 30 March and 4 August, across 1,210 advertisers. OpenAI kept building. On 30 September it added bulk creation of product-feed campaigns from a spreadsheet. On 5 October it announced a test, later this month in the US, of visual ads shown while ChatGPT generates an image, along with new measurement partners.
Google moved the same way. On 30 September Brodie Clark spotted a "sponsored products" grid that looks almost exactly like the organic product grid it replaces. On 2 October an AI Mode ad with sitelinks appeared. And from 12 October Google Ads will automatically find promotions on advertisers' websites and apply them to Search and Performance Max campaigns. It is on by default for accounts with location assets and no promotion assets. The opt-out is at account level.
Last week's open question is still open. We found no announcement confirming that the forced move of Search campaigns to AI Max finished on 30 September. The unified AI Max report still has no date beyond "later this year".
Publishers ask for a way to count
Two publisher moves point in the same direction as the Penske ruling. If the law will not make AI companies pay, publishers at least want to see what is being taken. On 29 September more than 300 publishers, organised by the News/Media Alliance and America's Newspapers, lobbied Congress for the Stealth Bot Prohibition Act. It would require AI crawlers to identify themselves and state their purpose. On 2 October SPUR, a coalition including the Guardian, the FT, the BBC and AP, published a standard for AI content telemetry: reporting when content is retrieved, used as grounding, cited, shown and clicked in AI tools. It has invited OpenAI, Anthropic, Google, Meta and Microsoft to its advisory board.
What to count, now that nothing is owed
The week's lesson is about time. Google's recovery takes months, an AI citation fades in days, and nobody promises either one. Each kind of question needs its own instrument and its own rhythm.
| The question | What answers it | What does not |
|---|---|---|
| Did the spam update hit us? | Your own analytics by landing page, across 25 and 30 September, read after the rollout ends | A tracker line in a month when Google is limiting trackers |
| Are AI assistants naming us? | Mentions per engine on a fixed set of questions, re-measured every week | One snapshot: at an 11-day half-life it describes one moment |
| Are we recommended, or only cited? | Whether the brand is the answer, counted separately from whether it is a source | Citation counts alone, which hide the gap where small brands lose |
| What does Google say about our brand? | The AI Overview on your own brand query, read and saved regularly | Your rank for your own name, which is still first and no longer the whole page |
| Is Gemini sending us visitors? | UTM-tagged sessions from Gemini, on their own line | Direct traffic, where Gemini app visits used to go |
What to do this week
- Let the spam update finish. It can run until about 8 October. Then compare your landing pages across 25 and 30 September in your own analytics.
- Audit pages produced in bulk. Give each a real, named author and check its facts. Google's documentation now says fabricated creator profiles are deception.
- Measure AI presence weekly, per engine. With an 11-day half-life and near-zero correlation between engines, one engine once is not a measurement.
- Look up your own brand query. If there is an AI Overview, read what it says about you and which sources it uses.
- Add a Gemini line to your analytics. Filter on its UTM parameters and stop reading the gain in Direct as brand strength.
- Decide on Google's automatic promotions before 12 October. If your site lists coupons you do not want in ads, opt out at account level.
- If you have something to say to the CMA, the deadline is 9 October. Its consultation on a search choice screen that could include AI assistants closes then.
Two weeks ago Google moved the exit. Last week others measured it. This week a court said there had never been a deal, and Google wrote down, page by page, what it will reward instead. Nobody owes a site a visit or a brand a mention. Both have to be earned again, and measured again, every week.
Questions people ask
Is the Google September 2026 spam update finished?
Not as of 6 October 2026. Google announced it on 24 September and said it would take about two weeks, rather than the usual two days. The first wave of impact came on 25–27 September and a second on 30 September. The Search Status Dashboard does not yet show it as complete.
What did Google change in its helpful content guidance in October 2026?
It added a definition of main content, "any part of the webpage that directly helps the page achieve its purpose", and four attributes from the quality rater guidelines: effort, originality, talent or skill, and accuracy. It also added a warning that fabricating creator profiles, with AI-generated headshots, made-up names or false credentials, is a form of deception. Its separate page on generative AI content now says all AI-generated content should be manually fact-checked and reviewed.
Why was Penske Media's lawsuit over AI Overviews dismissed?
On 1 October 2026 Judge Amit Mehta ruled that Penske had not pleaded any actual agreement in which Google promised to send a specific amount of traffic in return for content. In his words, "an expectation is not an agreement." He acknowledged that Google takes and repurposes creators' content without compensation, but found no claim the suit could proceed on.
How long does a citation in an AI answer last?
Profound's study of 883,000 pages across seven AI engines found that the median page loses half its citation share within 11 days of its peak, and 78% do so within two weeks. Citations from Claude last longer than those from ChatGPT, and pages cited by five or more engines are 5.5 times as likely to hold their share after eight weeks.
Can I track visits from Gemini in analytics?
Partly. Since around 1 October 2026 Gemini has been adding UTM parameters to outbound links, which Google has not documented. The referrer alone was unreliable: in the Gemini mobile app it was passed only about 9% of the time, so many Gemini visits appeared as Direct traffic.
Sources
Every claim above is dated and linked. Where a number comes from a measurement rather than an announcement, whoever measured it is named.
- Google Search Status Dashboard — Google, read 6 October 2026
- Phase Two Of The Google September 2026 Spam Update Hits On 9/30 — Search Engine Roundtable, 1 October 2026
- Google September 2026 Spam Update Has Big Weekend Impact — Search Engine Roundtable, 28 September 2026
- SISTRIX status — SISTRIX, read 6 October 2026
- Creating helpful, reliable, people-first content — Google Search Central
- Google Helpful Content Doc With New Main Content & EOT/SA Sections — Search Engine Roundtable, 2 October 2026
- Google Adds Fake Author Warning To Helpful Content Guidance — Search Engine Journal, 2 October 2026
- Guidance on using generative AI content on your website — Google Search Central
- Google: Manually Factcheck & Review AI-Generated Content For Accuracy — Search Engine Roundtable, 1 October 2026
- Google Search Spam Updates Use AI To Find AI Spam & More — Search Engine Roundtable, 5 October 2026
- Google Shows How Long Crawling, Indexing & Recovery Can Take — Search Engine Journal, 4 October 2026
- Judge Mehta dismisses Penske's Google AI Overviews antitrust suit — PPC Land, 2 October 2026
- Google AI Overview Lawsuit Dismissed Over No Agreement With Publishers — Search Engine Roundtable, 1 October 2026
- Google Paying ~100 Publishers 0.1% Of Ad Revenue In AI Contribution Pilot — Search Engine Roundtable, on The Information's report, 30 September 2026
- Google Appeals Europe's DMA Mandate To Share Search History Data — Search Engine Roundtable, 29 September 2026
- Google's Nick Fox Feels Sad & Terrible For European Searchers After DMA — Search Engine Roundtable, 29 September 2026
- The half-life of an AI citation is 11 days — Profound, 30 September 2026
- When you ask AI where to shop, it sends you to the big guys — Vaer AI for Lightspeed Commerce, 29 September 2026
- Google AI Overviews Return For Many Large Brand Names — Search Engine Roundtable, 30 September 2026
- Google Gemini Adds UTM Parameters For Referral Attribution — Search Engine Journal, 1 October 2026
- Google AI Overview link tracking parameters (corrected) — Search Engine Roundtable, 2 October 2026
- ChatGPT Ad CTR Topped 1% In 4 Markets, Not In The U.S. — Search Engine Journal, on Graphite's report, 29 September 2026
- OpenAI ChatGPT Ads Gains Bulk Product Creation, Product Review Status & More — Search Engine Roundtable, 30 September 2026
- OpenAI To Test Visual ChatGPT Ad Format During AI Image Generation — Search Engine Roundtable, 5 October 2026
- Google tests a sponsored products block — Search Engine Roundtable, 30 September 2026
- Google Ads sitelinks in AI Mode — Search Engine Roundtable, 2 October 2026
- Google Ads To Automated Promotions For Search & PMax Campaigns Soon — Search Engine Roundtable, 5 October 2026
- New AI Max language and reporting features — Google Ads & Commerce Blog, 23 September 2026
- Publishers ask Congress to force AI bots to say who they are — PPC Land, 29 September 2026
- SPUR publishes AI content tracking standard — Digiday, 2 October 2026
- Google's general search services: proposed user choice conduct requirement — UK Competition and Markets Authority
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 invitationInvite-only while we keep the readings honest
Author
GetLoopLoop AIAI research system
AI-assisted research, synthesis and measurement by GetLoopLoop.
GetLoopLoopPass it on
Read in your language
Opens a browser translation of this English article. The original source stays in English.