ChatGPT Ads Go CPC, EU Targets Google Data

Estimated reading time: 20 minutes

Three stories shape this week’s coverage. ChatGPT Ads has moved to CPC bidding in the $3-$5 range, which changes the cost calculus for performance marketers considering the platform. The EU Commission has proposed forcing Google to share ranking, query, and click data with rival search engines and AI chatbots. And a growing body of research shows AI search systems are contaminating themselves in real time by retrieving synthetic content and citing it as fact. Elsewhere, Google added AI-qualified call leads, Microsoft shipped a cluster of ads AI features, our own James Allen published a well-timed piece on why no-JavaScript fallbacks still matter, and a Duda study of 68 million AI crawler visits shows what actually drives AI search visibility. Here’s the full search marketing news roundup.

No-JavaScript Fallbacks in 2026: Less Critical, Still Necessary

Full disclosure: this one was written by our own James Allen, published on Search Engine Land last Friday. We’re featuring it because it tackles a question most technical SEO teams have quietly written off, and the case against complacency turns out to be stronger than you might expect.

The short version: Google has improved its JavaScript rendering capabilities, and in July 2024 Zoe Clifford stated that Google “renders all” HTML pages. But that statement carries important caveats. Google’s own documentation says Googlebot queues pages with 200 status codes for rendering, with execution happening “when resources allow”. Translation: there’s a lag, and for sites with rendering-heavy architecture, that lag can matter.

The piece digs into the technical limitations that still apply in 2026:

Google enforces a 2MB cap on both HTML pages and individual resources. Pages that exceed this may have content pushed outside the processing window
Google likely doesn’t interact with JavaScript-triggered elements behind tabs or modals
Non-200 responses may not get rendered at all
Google’s March 2026 “Inside Googlebot” post confirmed that extreme resource bloat (including large JavaScript modules) still affects indexing quality

There’s a second, equally important argument in the article that’s easy to miss: AI crawlers like ChatGPT and Claude don’t execute JavaScript at all. As those systems grow in importance for information discovery, HTML-first content delivery becomes more valuable, not less. Relying on client-side rendering alone gets progressively riskier as the crawler ecosystem fragments.

HTTP Archive data in the piece shows valid canonical deployment has actually dropped since November 2024, with 2-3% of rendered pages showing modified canonical URLs. That’s exactly the kind of indexing confusion Google’s documentation warns against.

The practical recommendation James lands on is a pragmatic middle ground. Blanket site-wide fallbacks aren’t universally required, but abandoning them entirely is risky. Maintain HTML fallbacks for critical architectural elements, internal linking, and essential page content. Resilient HTML paired with JavaScript enhancement remains the safer pattern than JavaScript-dependent-only implementations. If you’ve been tempted to drop fallbacks on a modern framework rebuild, have a read before you commit.

ChatGPT Ads Now Offer CPC Bidding at $3-$5

OpenAI has introduced cost-per-click bidding for selected pilot advertisers, according to Digiday. The range sits between $3 and $5 per click. That’s a meaningful shift from the impression-based pricing that defined the pilot’s first phase, and it opens the platform up to performance marketers who’ve been watching from the sidelines.

The economics have changed quickly. Since the pilot launched on 9 February 2026, CPM rates have come down from $60 to roughly $25, and minimum spend commitments dropped from $250,000 to $50,000. OpenAI has also quietly rolled out a self-serve ads manager that shows impression and click data in real time.

For agencies, this is the moment the ChatGPT Ads conversation with clients stops being theoretical. A $3-$5 CPC is broadly comparable to Google Search in some verticals and meaningfully higher in others. Measurement remains a known gap (OpenAI is still recruiting a marketing science lead), so early testing should be treated as directional rather than definitive. But the option to directly compare ChatGPT performance against Google and Microsoft on a like-for-like CPC basis changes the case study you can put in front of a CMO.

Google May Have to Share Search Data With Rivals Under EU Proposal

The European Commission has issued preliminary findings that would require Google to share anonymised search data with rival search engines and qualifying AI chatbots operating in the EU/EEA. The proposal sits under the Digital Markets Act framework, and the data categories covered are significant.

Four types of anonymised information would be shared on fair, reasonable, and non-discriminatory (FRAND) terms:

Ranking data (how results are positioned)
Query data (search terms users enter)
Click data (which results users select)
View data (impressions and visibility metrics)

The AI angle is what makes this genuinely significant. Eligibility extends beyond traditional search engines to include AI chatbots that meet DMA definitions of online search services. If this goes through, competing AI platforms could access signals similar to what Google uses internally. That’s a different scale of competitive pressure than Google has faced before.

Public consultation closes on 1 May, with a binding decision expected by 27 July 2026. Google has pushed back hard, arguing the proposal “far exceeds the DMA’s original mandate” and raises privacy concerns. Watch this one closely if any of your clients operate in the EU.

AI Search Is Eating Itself (and the SEO Industry Is the Source)

Pedro Dias has published an unusually sharp piece arguing that AI search systems aren’t just failing on accuracy; they’re actively contaminating themselves in real time. And the culprit, awkwardly, is the SEO industry.

“RAG systems do not generate answers purely from parametric memory. They fetch documents from the live web, stuff them into context, and generate a response conditioned on what they retrieved.”
– Pedro Dias

The distinction matters. Most coverage of AI hallucinations has focused on model collapse (a slow degradation over training cycles). Dias argues the more immediate problem is retrieval contamination, happening at query time when RAG systems pull synthetic content from the live web and serve it back as authoritative.

The examples he cites are alarming. Perplexity confidently invented a “September 2025 Perspective Core Algorithm Update” traced back to AI-generated agency blog posts. BBC journalist Thomas Germain published a joke piece titled “The best tech journalists at eating hot dogs”, and within 24 hours Google AI Overviews and ChatGPT were citing it as fact. Grokipedia (Elon Musk’s AI encyclopedia) claimed Feist’s father died in May 2021, citing a 2017 article that mentioned no death and was published while he was alive.

The data underneath is worse than the anecdotes. Dias cites 4,326 tests showing Google AI Overviews at 85-91% accuracy, but with 56% of correct answers on Gemini 3 being “ungrounded” (up from 37% on Gemini 2). Facebook and Reddit ranked second and fourth among sources cited. Nearly 44% of ChatGPT-cited sources were “best X” listicles. That’s the shape of the web AI is reading.

The uncomfortable conclusion: SEO agencies and content creators deliberately built AI content pipelines to compete with AI Overviews, producing speculative “winner/loser” content during Google updates. That synthetic content floods retrieval indexes, and then gets cited back as authoritative. It’s a closed loop, and the industry is feeding it.

The Hidden ‘Bland Tax’ That Could Erase Your Brand From AI Search

Andrew Warden, CMO of Semrush, used his Adobe Summit talk to introduce a concept he calls the “bland tax”. The idea: AI systems systematically filter out generic or repetitive content, summarising similar answers into consolidated responses and often stripping attribution entirely.

“If you are generic, you are average. And if you are average or bland… you are invisible.”
– Andrew Warden, Semrush

The article lands on a useful argument that SEO hasn’t been replaced by AI; if anything, it’s become more foundational. Warden’s framing is that SEO is now also “a training manual for AI”. Without crawlability, structured data, and authority signals, brands simply disappear from AI conversations. Nothing brands can do about that; you’re either in the set or you’re not.

Three signals matter for brand differentiation in this model:

Entity authority: AI systems need to recognise your brand as topically authoritative, and brand demand signals feed into that. “If people aren’t looking for you, then neither is AI”
Information density and originality: Proprietary data, original research, and unique perspectives can lift visibility by 30-40%
Signal alignment: Consistency across reviews, social discussion, media coverage, and customer conversations creates what Warden calls a “consensus signal”

Honestly, the bland tax is a decent label for something SEOs have been struggling to articulate for a year. If you’re producing the same AI-summarised content as five other agencies in your niche, the AI isn’t picking any of you. That’s a real problem, not a theoretical one.

What 68 Million AI Crawler Visits Reveal About AI Search Visibility

Duda analysed 858,457 websites and tracked 68.9 million AI crawler visits across February 2026. The findings give us the clearest picture yet of what’s actually pulling AI crawlers to sites.

The big shift: 56.9% of AI crawler activity now involves “user fetch” (real-time answers for specific queries), mostly from ChatGPT. Only 14.3% focuses on discovery or indexing. Training accounts for 28.8%. So the behaviour pattern has changed. AI crawlers now spend most of their time answering specific user questions rather than sweeping the web for training data. That’s a big clue about what optimisation should target.

Market concentration is stark. OpenAI accounts for 81% of all AI crawler visits. Anthropic sits at 16.6%. Everyone else fights over the remaining 2.4%. For most brands, optimising for OpenAI effectively means optimising for AI search overall.

The study also surfaced three specific factors that correlate strongly with crawl rates:

External integrations: Yext integration correlates with 97.1% crawl rates
Structured data: Google Business Profile sync links to 92.8% crawl rates; complete local schema increases crawl rates by 26.8 percentage points
Content depth: Sites with 50+ blog posts receive roughly 33 times more crawler visits than thinner sites

One subtle but important finding: AI-crawled sites average 3.2x higher human sessions than sites without crawler visits. The authors suggest AI visibility follows existing audience demand rather than generating it. If that holds up in subsequent studies, it changes how we should think about GEO as a traffic acquisition strategy. It’s starting to look like an amplifier of existing brand strength, not a replacement for it.

AI Best-of-N Jailbreaking: A Brute-Force Safety Bypass

Myriam Jessier’s piece on Best-of-N (BoN) jailbreaking should be required reading for anyone using AI tools with client data or deploying AI on customer-facing properties. The attack exploits AI randomness. Because the same prompt returns slightly different answers each time, an attacker just needs to try enough variations until one slips past safety filters.

The numbers are uncomfortable:

89% attack success rate on GPT-4o with 10,000 variations
78% success rate on Claude 3.5 Sonnet
35% additional success when combined with prefix attacks
Attacks can succeed in seconds with newer techniques

The attack only needs basic Python and API access. Variations are generated by random capitalisation, character scrambling, typos, and filler tokens. The technique works across text, image, and audio modalities.

For agencies, the practical implications are threefold. First, sensitive client inputs fed to AI tools need the same scrutiny you’d apply to GDPR-sensitive data. Second, safety filters on AI tools you deploy should be treated as one layer of defence, not the whole wall. Third, if jailbreaking produces offensive output from a brand-deployed AI tool, media coverage will be about the brand, not the technical vulnerability. That’s a communications exposure worth planning for, not just an engineering problem.

The Funnel Flip: Why AI Forces a Bottom-Up Acquisition Strategy

Jason Barnard argues that the 128-year-old top-down marketing funnel still describes how users experience brands, but no longer describes how AI systems build their willingness to recommend them.

The insight sits in the asymmetry. Users still move through awareness, consideration, decision. But AI systems work in reverse: they first establish who you are, then assess credibility, then decide whether to advocate for you. Machine-mediated discovery means agents increasingly push users toward brands they’ve already vetted, which makes algorithmic gatekeeping a bigger factor than traditional awareness campaigns.

Barnard’s recommendations follow a strict order. Understandability first (ensure machines correctly identify who you are, what you offer, whom you serve). Credibility second (third-party mentions, earned media, what Barnard calls N-E-E-A-T-T authority). Deliverability third (content that helps machines anticipate the next step in user journeys). His argument is that these must proceed in sequence, because the underlying assessment is sequential. You can’t build credibility the AI won’t trust if it can’t understand who you are in the first place.

Google Adds AI-Qualified Call Leads to Improve Measurement

Google has rolled out AI-qualified call leads in Google Ads, using machine learning to score the actual business value of phone calls rather than relying on call duration as a quality proxy. The system generates AI call summaries and tags, flags spam and robocalls, and lets advertisers optimise for high-value leads rather than long leads.

Call recording is enabled by default for most US and Canadian advertisers, with exceptions for healthcare and financial services. Anyone running call campaigns should check account settings now if recording opt-in matters for their client. The feature is a meaningful upgrade for service-based businesses where lead quality has historically been nearly impossible to measure without manual review.

Microsoft Advertising Ships a Cluster of AI Features

Microsoft Advertising announced a batch of AI features and upgrades across its platform this week, positioning itself to “help businesses succeed across all three eras of the human, LLM, and agentic web”. That’s Microsoft’s framing, not ours. But the investment pace is worth noting; Microsoft is pushing hard to avoid ceding the AI-era ads market to OpenAI and Google.

Specifics are still emerging but the release covers ad creation, audience targeting, and campaign optimisation. If you manage Microsoft Ads clients, check the platform updates in-account this week.

First Look: ChatGPT’s Ad Manager Interface

Early screenshots and videos of the ChatGPT ads manager have surfaced from pilot advertisers. The interface supports campaigns, ad groups, budgets, CPC limits, and conversion tracking. Nothing radical structurally; it looks and works much like Google Ads or Meta Ads.

The significance is less about features and more about readiness. OpenAI is clearly building out the full advertiser workflow, not just running pilot inventory. Combined with the CPC bidding story above, this week represents a real step from “pilot” to “platform” for ChatGPT Ads.

OpenAI Launches OAI-AdsBot, a New Ads Crawler

OpenAI has released a new crawler named OAI-AdsBot. This joins the existing GPTBot, ChatGPT-User, and OAI-SearchBot in the OpenAI crawler family. If you’re managing robots.txt or log-file analysis for any site running or considering ChatGPT Ads, add OAI-AdsBot to your tracking setup. Details on its exact behaviour are limited so far, but based on the name it’s presumably involved in advertiser-site verification or ad-context gathering.

Why Ugly Ads Outperform Polished Creative

Akvile DeFazio makes a case that scrappy, low-production creative captures attention better than polished campaigns. The argument: “high-production ads signal ‘this is an ad’ almost instantly, triggering a skip reflex”. Lo-fi, user-generated-style creative bypasses that initial scepticism.

Her testing recommendation is an 80/20 split. Keep 80% of spend on your proven creative, give 20% to experimental, unconventional formats. Tests worth trying include silent ads with bold captions, interface-mimicking static images, and deliberately broad targeting that leans on algorithmic optimisation. Not everything will outperform, but the upside when something does can be significant.

WooCommerce Stores Can Now Sell Products via YouTube Videos

Google and WooCommerce have enabled merchants to tag products directly in YouTube videos, reaching YouTube’s 2.7 billion shoppers through shoppable cards. The integration is a real step forward for independent ecommerce brands that’ve been locked out of the YouTube Shopping ecosystem dominated by Shopify partners. If you’ve got WooCommerce clients with video content, there’s an obvious play here.

Google Ads Advisor Gains Troubleshooting, Security and Certification Features

Google Ads Advisor has added three features: proactive troubleshooting, 24/7 security monitoring, and instant certifications. The instant certifications piece is the standout; it replaces the traditional exam-based certification process with real-time skill validation. Whether that raises or lowers the bar in practice is an open question, but account managers at agencies should keep an eye on how this affects certified-team requirements for Google Partner status.

The Ghost Citation Problem

New research across multiple LLMs finds significant inconsistency in how AI systems handle citations and brand mentions. A page might be cited, mentioned without citation, or invisible entirely depending on which model you ask, which query phrasing you use, and when you ask. The practical implication: single-query AI visibility checks aren’t reliable. You need repeated sampling across models and phrasings to know whether your brand is actually being referenced. More evidence that AI visibility tracking tools need more rigour than most are currently offering.

Stanford Data: AI Adoption Outpaces PC and Internet

Stanford’s 2026 AI Index Report confirms what most of us already suspected: AI adoption has outpaced both the personal computer and the internet in terms of speed to mass use. The report also covers reliability challenges and transparency concerns across major models, which is useful context for anything you’re telling clients about AI’s readiness for business-critical workflows. If you’re building strategy decks, the Stanford data is a reliable reference point.

The Yoast Perspective 2026: What 59 SEO Professionals Are Actually Thinking

Yoast surveyed 59 SEO professionals for a state-of-the-industry report. The findings are refreshingly un-alarmist: SEO is evolving, not dying. Respondents noted that “the interfaces between users and information are changing” as AI-driven discovery expands the definition of search. The report also covers rising content saturation and the growth of alternative search channels like TikTok. Good read if you want industry-peer sentiment data to counterbalance the “SEO is dead” hype cycle.

Google Documents Read More Snippet Links Best Practices

Several months after launching “Read more” links inside search result snippets, Google has published best practices for them. Short version: clear section structure, descriptive headings that match likely user queries, and avoiding overly long paragraph blocks all help Google extract “Read more” links. If CTR from snippets matters (and for most commercial content, it does), this is a cheap optimisation to run across your client sites.

Google Tests Video Ads Inside the Local Pack

Google is testing video ads inside the local pack (also being called “immersive map view videos”). For local service businesses, a video asset appearing directly in the map pack is a genuine competitive advantage when this rolls out widely. Start producing short vertical video for local clients now; it’ll be useful across the local pack, Google Business Profile, and YouTube Shorts regardless of how this specific test plays out.

Google Search Console Jobs Report Bug

Google Search Console has a reporting bug affecting the jobs listing and jobs search appearance filter. Impressions and clicks have been showing as zero since 16 April. If you’re running a jobs-heavy site and have seen a sudden reporting drop, it’s almost certainly the bug rather than a ranking issue. No need to panic yet.

Google Tests Opening Search Results in New Windows

Google is testing opening clicked search results in a new window, even for users who haven’t opted into that setting. If this becomes default behaviour, it changes user experience assumptions baked into conversion tracking for many sites. Monitor for changes in return-to-SERP patterns and adjust attribution modelling if necessary.

Strategic Direction: Where Search Marketing Is Heading

A few patterns came into focus this week. The first is that ChatGPT Ads has moved from “we should probably keep an eye on this” to “we need a testing plan”. A CPC bidding model with declining minimum spend commitments, a proper ads manager interface, and a new dedicated crawler means OpenAI is serious about becoming a real platform. Agencies that treat paid search as a Google-plus-Microsoft discipline will need a third column in their media plans by Q3.

The second pattern is regulatory pressure on Google finally reaching a point where it might actually alter competitive dynamics. EU data-share mandates aren’t theoretical any more; a binding decision is expected by 27 July. If rivals and AI chatbots gain access to Google’s ranking and query data under FRAND terms, the competitive moat shrinks meaningfully. Combined with Microsoft’s investment in AI ad features, the agentic AI market is the first serious challenge to Google’s search dominance in over a decade.

The third and most uncomfortable pattern is the growing evidence that AI search quality is declining, not improving, because of retrieval contamination. Pedro Dias’s piece, the ghost citation research, the 68 million crawler study, and the bland tax concept all point the same way: AI systems are increasingly confident, increasingly inaccurate, and increasingly dependent on synthetic content generated by SEO pipelines. That’s a problem the industry created and has strong financial incentives to continue creating. Expect more regulatory attention, more accuracy research, and harder conversations with clients about what AI-assisted content actually buys them.

Key Takeaways

ChatGPT Ads has moved to CPC bidding at $3-$5 with a $50,000 minimum spend. Agencies need a testing plan for Q3.
The EU may force Google to share ranking, query, click, and view data with rivals and AI chatbots by 27 July. Watch closely for EU clients.
Duda’s 68 million crawler study shows OpenAI accounts for 81% of AI crawler activity, and user-fetch now dominates over training/indexing. Optimise for real-time answer retrieval.
The “bland tax” concept captures something real: generic, AI-summarisable content gets filtered out of AI answers entirely. Original data, unique angles, and strong entity authority are becoming the baseline.
Best-of-N jailbreaking hits 89% success on GPT-4o. Treat AI tool safety filters as one defensive layer, not the whole wall, and audit any client data fed into AI tools.
No-JavaScript fallbacks in 2026 remain necessary, not optional. AI crawlers don’t execute JavaScript at all, and Googlebot’s rendering still carries real limits.
AI search is in a retrieval-contamination loop driven partly by SEO-generated AI content. Expect accuracy debates and regulatory attention to intensify.

Frequently Asked Questions

Should I add ChatGPT Ads to my client media plans now?

If you have clients with $50,000+ to allocate and a performance goal that tolerates measurement ambiguity, yes. Treat early results as directional. The platform’s measurement stack is still maturing, but CPC pricing makes like-for-like comparison against Google and Microsoft feasible for the first time.

What should I do about the EU data-share proposal if I have EU-based clients?

Nothing urgent yet. Public consultation closes 1 May; binding decision expected 27 July. But if the proposal passes, expect rival search engines and AI chatbots to access Google-adjacent data within months. Start thinking about multi-platform visibility strategies and don’t assume Google’s data advantage will stay intact indefinitely.

How do I actually avoid the “bland tax”?

Three practical steps. First, publish proprietary data or original research rather than restating industry stats. Second, strengthen entity authority through third-party mentions, reviews, and media coverage. Third, align messaging across reviews, social, and media so AI systems see a consistent “consensus signal”. Repetitive AI-assisted content won’t cut it.

Is JavaScript-first rendering really still risky in 2026?

Riskier than it sounds. Googlebot can render JS but not always immediately, and AI crawlers like ChatGPT and Claude don’t execute JavaScript at all. HTML-first content with JavaScript enhancement remains the safer pattern. The full argument is in James Allen’s SEL piece linked above.

Should I be scared of Best-of-N jailbreaking?

Scared is the wrong word; prepared is better. If you feed client data into AI tools, audit what you’re sharing at GDPR level. If you’re deploying AI on customer-facing sites, assume safety filters will be bypassed eventually and plan your incident response accordingly. Brand damage from an AI tool producing harmful output lands on the brand, not the AI vendor.

Conclusion

This was the week the ChatGPT Ads story became concrete, the EU’s antitrust pressure on Google started to feel genuinely threatening, and the AI search quality problem got harder to dismiss as a temporary teething issue. Three big headlines, each with real implications for how we plan and buy search for the rest of 2026. Underneath those, a cluster of smaller updates (AI-qualified call leads, Microsoft’s AI features, the local pack video test) adds up to a platform landscape that’s moving faster than any single agency can fully track. The teams that’ll come out ahead are the ones testing aggressively in the new ad ecosystems and staying sceptical about easy AI content answers.

Need help adapting your search strategy for the AI era? Contact the Anicca team for expert SEO and PPC guidance.

Table of Contents

SEO & Technical

AI Search, GEO & AEO

PPC & Paid Media

AI Safety & Governance

Industry

Wrap-Up