Search Signal #02
Your pricing page surfaces in 46% of AI answers but leads just 12% of the time. Plus: Search Console's AI report goes global, and the DOJ backs OpenAI.
Vikas Kalwani
Founder, SAASY LINKS
01 — Your pricing page surfaces, then Vendr and Reddit get cited
Profound and Growth Unhinged collected 7,600+ AI responses to pricing queries across 77 Cloud 100 companies and six engines, between 29 July and 10 August. Company-owned pricing pages surfaced in 46% of responses but were cited first only 12% of the time. Filling the gap: Vendr at 18.7%, Reddit at 18.6%, G2 at 15.9%. ChatGPT was the most generous to brands at 38%.
Why it matters: When a prospect asks an AI what you charge, the answer usually comes from somewhere you do not control. One company in the sample saw negative pricing language in 72 of 76 responses.
Action: Ask ChatGPT, Perplexity and Google AI Mode what your product costs, three times each. Note which source gets cited first. If it is Vendr, Reddit or G2, that page is your pricing page now, and fixing it is an outreach job.
02 — Search Console’s AI performance report is now global
Google finished rolling out the generative AI performance report on 31 August, three months after starting. It sits as an expandable tab inside Performance, showing how often your URLs surface across AI Overviews, AI Mode and AI Overviews in Discover, broken out by impressions, pages, countries, devices and dates. Google’s own wording: it “does not include click data”. The paired opt-out toggle keeps content out of those features entirely.
Why it matters: This is the first Google-native AI visibility number your clients can pull without you. Impressions with no clicks is also the most honest description of the surface anyone has published.
Action: Open Search Console, go to Performance, and expand the AI tab before your next client call. Get in front of the framing rather than explaining it after they find the tab themselves.
03 — Product pages take 24.1% of B2B AI citations, forums take 4.2%
Ten Speed analysed 7,387 citation appearances across 170 evaluation-stage B2B prompts, run through ChatGPT, Perplexity, Claude and Gemini. Product pages took 24.1%, articles and blogs 17.4%, comparison content 13.3%, listicles 13.2%, third-party profiles such as G2 and Capterra 7.2%, discussion and community 3.7%, video 0.5%. Brand-controllable content totals 88.3%. Comparison prompts were the efficiency outlier: 20% of prompts produced 26.7% of citations.
Why it matters: This cuts hard against the “optimise for Reddit” consensus, at least at evaluation-stage intent. But it is agency-published, single point in time, never significance-tested, and the authors say so.
Action: Do not put 24.1% in a deck as fact. Do run the same test on your own accounts: 20 evaluation-stage prompts, log what gets cited, compare against this split.
04 — The US government told a court that AI training is fair use
The United States filed a Statement of Interest on 1 September in the consolidated OpenAI copyright case, the first formal federal position on the AI copyright docket. The DOJ argued training is “exceedingly transformative”, called a licensing mandate “legally incorrect”, and warned it would “concentrate AI development among a few firms”. Three days later OpenAI moved for summary judgment, reporting 24 instances of verbatim reproduction across 20 million ChatGPT conversations from 10.8 million asserted articles.
Why it matters: The strongest tailwind AI companies have had in this litigation, and it came from the government rather than the court. If training settles as fair use, publishers keep leverage only at the retrieval and citation layer.
Action: No immediate action, but stop building strategy on the assumption that licensing deals will force AI companies to pay for content. Plan for the citation layer being the only lever.
05 — Your AI visibility tracker is manufacturing its own plateau
A new paper ran 50 buying questions across six engines, 15 runs each, using open extraction rather than a fixed brand roster. Five of six engines were still surfacing brands never seen before at run 15 in 86 to 92% of cells. A single run captures only 62 to 77% of the five-run brand set. The sharpest finding: fixed-roster extraction, which is what most commercial visibility trackers do, produces flat saturation curves on the identical responses.
Why it matters: If your dashboard tracks a preset competitor list, it shows a stable picture the underlying responses do not support. Six pages, single author, no stated affiliation, so treat the mechanism as the finding rather than the numbers.
Action: Never report a single-run visibility figure as the number. Run each tracked prompt at least three times and report the range.
06 — A model swap stopped AI Mode citing anyone, and Google called it a bug
Google put Gemini 3.8 Flash into AI Mode on 2 September, less than three weeks after 3.7. Within a day, Gagan Ghotra and Glenn Gabe documented responses returning far fewer links than the default model, in many cases none at all. Robby Stein, VP of Product for Search, replied publicly: “This isn’t working as intended, and we’ll roll out a fix soon.” Links were back the following morning.
Why it matters: A routine model swap silently removed the citation surface for about a day, and it took practitioners with screenshots to catch it.
Action: Keep a dated log of model releases alongside your AI visibility data. Without it you will attribute a platform-side dip to a client’s content and spend a month fixing the wrong thing.
07 — ChatGPT is now legally a search engine in the EU
The European Commission designated ChatGPT a Very Large Online Search Engine under the Digital Services Act on 31 August, the first time an AI chatbot has been classified this way. The trigger: OpenAI declared 159.1 million average monthly EU recipients for ChatGPT search, against a 45 million threshold. Reddit and Roblox were designated as Very Large Online Platforms in the same decision. Compliance is due by January 2027.
Why it matters: The vetted-researcher data access obligation is the sleeper. For the first time an AI answer engine may have to hand over data this field currently estimates from small samples.
Action: Stop calling AI search an emerging channel in client decks. 159.1 million EU monthly users is a channel.
08 — ChatGPT Shopping flipped from web search to merchant feeds in one day
Profound tracked ChatGPT Shopping-mode product recommendations across 687 customers between 1 July and 24 August. On 10 July, the share sourced from integrated merchant feeds rather than web search rose from 8.26% to 61.54% in a single day. Merchant concentration tightened with it: top-10 merchant share went from 22.5% to 41.8%, while unique merchants fell from 13,524 to 10,607. Shopify accounts for roughly 35% of feed-integrated retrieval.
Why it matters: Vendor research on Profound’s own paying customers, so the swing sizes are inflated. The mechanism holds regardless: for a commerce client, ChatGPT visibility is now decided by feed integration, not by anything you write.
Action: For any client selling products, check feed integration status before selling them content work. The content will not fix a missing feed.
09 — Indig on listicles: freshness dominates, self-promotion does nothing
Kevin Indig analysed 59,984 SERPs across 60,000 US queries in 15 verticals, plus 3,000 crawled listicles. Freshness was the strongest signal: 66.6% of top-3 pages carried dates within the previous two years, against 57.3% in positions 8 to 10, and old or unparseable dates carried 56% lower odds of ranking top 3. The widely-repeated claim that self-promotion inside a listicle drives growth did not survive controls: 4% stronger growth, p=0.91, indistinguishable from zero.
Why it matters: The best-controlled study of the week, and the null result is the useful half. If you buy or sell listicle placements, the honest pitch is presence and freshness, not the self-promotion mechanism everyone assumes.
Action: Audit your highest-value listicle placements for visible, genuinely updated dates. Republishing a stale placement with a current date and real edits is the cheapest win on this list.
10 — A judge refused to break up Google’s ad tech and rewrote the auction instead
On 2 September Judge Leonie Brinkema declined to order divestiture of AdX or the DFP publisher ad server, 17 months after finding Google illegally monopolised both markets with 91% publisher ad server share. The remedy is behavioural: real-time AdX bid data must be shared with rival ad servers, unified pricing rules are deprecated, first look and last look are prohibited, and differential revenue sharing based on ad server choice is barred.
Why it matters: Two structural-remedy attempts against Google have now landed on conduct rules rather than breakups.
Action: Plan on Google keeping its current shape. Stop pricing client strategy around a court-ordered reset that keeps not arriving.
11 — Brussels is asking publishers whether Google’s AI opt-out is a real choice
The European Commission’s antitrust arm sent publishers a questionnaire with a 28 August deadline, asking whether they intend to use Google’s opt-out from AI search features and what would drive that decision. Google announced the opt-out on 3 June, the same day the UK CMA imposed its Publisher Conduct Requirement, which requires opt-out “without harming search rankings”. The European Publishers Council filed a complaint in February over content used “without consent”.
Why it matters: The opt-out only means something if using it does not cost you rankings, and that is exactly what Brussels is testing. A finding that the choice is illusory would force separation of AI grounding consent from ordinary indexing.
Action: Do not use the AI opt-out yet. The toggle currently costs you impressions with no evidence it buys anything back.
12 — ChatGPT Ads passed $1bn annualised in under 200 days
OpenAI reported ChatGPT Ads reaching “$1 billion in annualized revenue run rate” in under 200 days, and opened self-service across India, Europe, the Middle East and North Africa, taking the platform past 40 countries. The ad stack is filling in fast: Custom Audiences now support over 5 million members, the Measurement Pixel accepts additional hashed customer data, and an Ads Manager plugin allows campaign management in natural language inside ChatGPT.
Why it matters: The paid side of ChatGPT is professionalising far faster than the organic citation side, which is the same asymmetry Google ran through the 2000s. Budget conversations get harder when a fully measurable paid channel sits next to an organic one reporting impressions and no clicks.
Action: If you run paid for any client, request access now. Being early on a channel this immature is worth more than being good on a mature one.
Also worth knowing
- Google’s ranking systems went quiet. No confirmed update and no incidents on the status dashboard all week, which is unusual after August’s volatility.
- The GEO detection arms race got two papers on the same day. GEO Defender reported cutting attack success from 50.32% to 6.20%. Counter-GEO-Bench reported that three off-the-shelf safety guardrails cut it by at most 5.7%, because safety taxonomies catch policy violations, not fluent misinformation.
- AI Mode shows different and pricier shopping results than the carousel. Productrise compared 2m+ listings and found only 1.28% overlap, with AI Mode more expensive 68.4% of the time, averaging 21.6% higher. Google said all results come from the same Shopping Graph.
- USA TODAY dissolved its audience team. SVP Monica Richardson: “search traffic is under pressure, and platforms are increasingly keeping user experiences to themselves.” Condé Nast expects search to fall to single-digit percentages of total traffic.
- Exact and phrase match keywords started serving ads in AI Mode. Previously the preserve of broad match with smart bidding. Google’s Ginny Marvin called it “a small experiment”.
- On the calendar. 15 September: Cloudflare’s training and agent crawlers get blocked by default on ad-supported pages for new domains. 17 September: Google begins distributing licensing agreements under the European Search Dataset Licensing Program, with AI chatbots explicitly in scope as licensees.
SAASY LINKS
We build the visibility this newsletter reports on
Editorial links, AI citations, and organic pipeline for B2B SaaS. Senior-led, 90-day sprints, no long-term contracts.
Book a Free Strategy Call