AI Mode Link Fixes, Bing Grounding, AI Bidding

Estimated reading time: 25 minutes

Search news cycled up sharply on 6-7 May after a quiet bank-holiday weekend. Google rolled out five fixes to how AI Mode and AI Overviews link to publisher pages. Microsoft published a technical defence of why Bing needs a separate AI-grounding index. Google Ads pushed Journey-Aware Bidding and demand-led pacing into beta. Add a Canadian musician’s $1.5M defamation case against Google over a false AI Overview, the official launch of OpenAI’s self-serve ChatGPT Ads, and a stack of smaller SEO and PPC updates, and there’s plenty of search marketing news to work through this week.

Google's Five New Link Improvements in AI Mode and AI Overviews

This might be the most important publisher-facing shift Google has shipped in 2026 so far. On 6 May, Google announced five distinct changes to how it links to web pages from inside AI Mode and AI Overviews, a direct response to the persistent complaint that AI summaries hoover up source content while burying source attribution.

The five changes, in plain terms:

Further exploration suggestions appended to AI responses, pointing at deeper analyses on the same topic.
Subscription labels marked next to news outlets a user already pays for. Google reports that in early testing, people were significantly more likely to click links once they were labelled as belonging to a subscription the user already held.
Community perspectives previewed inline, pulling in posts from forums and social platforms with attribution by handle.
Inline contextual links placed directly next to the relevant sentences in an answer, rather than collected at the end.
Desktop link previews on hover, showing the destination site name and page title so users have a clue what they’re about to click.

There are two reads on this. The optimistic one is that Google is finally treating the source link as part of the user experience, not an afterthought. The cynical one is that AI Mode click-through rates have been ugly enough internally that this was forced.

For SEO teams, the practical implication is that semantic clarity inside content (clear titles, descriptive H2s, factual paragraphs that hold up out of context) now affects whether your page gets selected for an inline link versus an end-of-answer suggestion. Schema.org Article and Subscription markup also matter more, especially the latter for any subscription-gated publisher.

The pattern across all five updates is the same: AI Mode is turning into something between an answer engine and a curated index, with citation logic that increasingly resembles an editor’s judgement about which sources strengthen which claims.

Bing Explains Why AI Grounding Needs a Smarter Index Than Search

Microsoft Bing published a technical blog post on 6 May making a public case for why grounding for AI answers cannot ride on top of a traditional ranking index. It’s the most explicit acknowledgement we’ve seen from a major platform that “search index” and “AI answer index” are diverging.

The argument boils down to one observation: traditional search lets the user self-correct by scanning ranked links and choosing what to trust, while a grounded AI answer commits to a single response. That commitment shifts what the index has to assess. According to Microsoft, an AI grounding index needs to know whether each candidate page:

Survives chunking and transformation without losing meaning.
Has a clearly identified source.
Is fresh enough to be acted on.
Contains facts that are actually retrievable as evidence.
Conflicts with other sources, before that conflict reaches the user.

From Krishna Madhavan, Knut Risvik and Meenaz Merchant on the Bing engineering team:

“Indexing for grounded AI answers is not a reinvention of search. It is a major evolution of it. Grounding commits to an answer.”

What’s actionable here for the Anicca client base is the freshness rule. Microsoft singled out stale content as a structural risk for AI answers in a way that’s stronger than the equivalent risk in classical search: stale ranks lower; stale grounds wrong. If you have evergreen pages that haven’t been touched since 2023 sitting in critical category positions, those are the pages most likely to misfire in AI answers and pull a confidence-weighted citation share down with them.

Ashley MacIsaac Sues Google for $1.5M Over Defamatory AI Overview

This one matters more than it looks. Canadian fiddler Ashley MacIsaac filed a defamation suit in Ontario Superior Court of Justice this past week, seeking at least $1.5 million in damages. The trigger: Google’s AI Overview reportedly described him as having been convicted of sexual assault, internet luring of a child, and assault causing bodily harm, and as appearing on a national sex offender registry. None of that is true. A First Nation organisation that had booked MacIsaac to perform reportedly cancelled the engagement after seeing the summary.

Why it matters beyond a single celebrity defamation case is the precedent. Google’s standard line on AI Overview errors has been that the summaries are “dynamic and frequently changing” and that mistakes feed system improvements. MacIsaac’s lawyers are framing that defence as inadequate. As MacIsaac himself put it: “This was not a search engine just scanning through things and giving somebody else’s story.” The argument turns on whether composing an answer constitutes publishing it in a legal sense, separate from the linked sources.

Section 230 (or its non-US analogues) was written for ranked links, not generated answers.

For brands and individuals, this is the first sizeable test of AI Overview liability and, if it goes against Google, expect the bar for what counts as adequate sourcing in AI answers to rise quickly. For SEOs, it’s a reminder to audit how your clients are described inside AI Mode and AI Overviews, particularly anyone with a common name shared with someone notorious. We’ve seen that pattern misfire on a couple of healthcare clients already.

Google Tests Web Bot Auth, a Cryptographic Crawler Verification Standard

What happens when “user-agent” stops being a credible way to tell good bots from bad ones? Google’s answer is a new experimental cryptographic protocol called Web Bot Auth, formally the HTTP Message Signatures Directory.

The mechanism is simple enough. A bot publishes its public keys at a /.well-known/ location in JSON Web Key Set (JWKS) format. Each request it sends carries a new “Signature-Agent” header that points at that key directory. The receiving server verifies the cryptographic signature before deciding whether to honour the request. In effect, the bot proves who it is rather than claiming who it is.

Why now? Because the population of LLM crawlers, AI scrapers, and synthetic bots impersonating Googlebot has grown beyond the point where IP allowlists and reverse DNS can keep up. Web Bot Auth gives publishers a path to a verified-bot whitelist that can’t be spoofed by changing a user-agent string.

The protocol is still experimental. Google isn’t signing every request yet, and publishers should keep using IP, reverse DNS, and user-agent verification alongside it, not in place of it. But if Web Bot Auth ships properly, expect server-side bot policies to get more interesting fast: imagine being able to grant ChatGPT access to your blog while blocking GPTBot’s training crawl, with a cryptographically-verifiable distinction.

AI Overviews Now Display Author Names for LinkedIn and Medium Sources

A small but instructive UX tweak from Google: AI Overviews started showing the full author name alongside the source for citations from LinkedIn and Medium, two platforms where the platform name alone tells you nothing about who you’re trusting.

It’s a sensible move (anyone can publish on LinkedIn or Medium, and the credibility of a citation depends entirely on the author), and a quiet endorsement that personal brand matters more than domain authority for these platforms specifically. If your client team is publishing thought leadership on LinkedIn rather than a corporate blog, the visible author name is now a citation signal.

We expect this pattern to spread. The same logic applies to Substack, X long-posts, and any platform where domain-level authority is decoupled from author authority.

Why Scaled AI Content Hits a Quality Wall

Dan Taylor at Search Engine Journal argued this week that the pattern of “AI content scales nicely for three months, then collapses” isn’t really an AI problem. It’s a content strategy problem that AI happens to make easier to commit at scale.

His core point is that Google gives new URLs a freshness boost during sampling. The crawler gives a content batch attention, watches user behaviour, and decides whether to keep allocating crawl budget. If the sample fails the user-engagement threshold, Google reduces resources across the whole batch, not just the offending pages. AI-produced volume content is especially exposed because its weakest pages tend to share signatures (thin entity coverage, generic structure, padding sentences) that get spotted in the sample.

Taylor’s framing:

“AI is simply the latest, and easiest, scapegoat for a fundamental breakdown in the content pipeline.”

His recommended shift is “from production scale to quality maintenance at scale.” In practice that means editorial oversight that catches the bottom-decile pages before publication, internal links that don’t dilute focus across thin variants, and a rolling refresh policy that touches the worst-performing 10% of content quarterly rather than just adding more.

If you’re commissioning AI-assisted content for clients, this is the strongest argument in print so far for budgeting human editorial review as a non-negotiable line item, not an optional polish step.

The Citation Is Not a Click: Forrester on AI Visibility ROI

How do you justify spending on AI search visibility when the click data looks tiny? Duane Forrester’s article at SEJ this week is the cleanest argument we’ve read for why most of the industry is asking the wrong question. His framing: a citation in an LLM answer is not “functioning as a routing instrument.” It’s a “grounding artifact” or “confidence hedge,” and trying to measure it like an organic search click obscures the actual mechanism.

Forrester’s data points are uncomfortable. News publisher traffic fell from 2.3 billion to 1.7 billion monthly visits between mid-2024 and May 2025, a 26% drop. Zero-click searches climbed from 56% to 69% over the same window. Meanwhile AI infrastructure capex from major platforms is committed at roughly $660-690 billion for 2026, almost double 2025, with Bank of America estimating that AI capex will consume 94% of operating cash flows across that period.

The framework he proposes replaces click-counting with three citation-quality measures:

Citation prominence (how often your brand is named as the grounded source).
Citation trustworthiness (whether the AI hedges with you or asserts with you).
Influence on the answer (whether the cited claim is decorative or load-bearing).

For agency-side measurement, that’s a meaningful reframe. We’re already finding clients want to see weekly tracking on which AI engines name them, with what context, and against which competitors. Forrester’s argument gives that work a defensible ROI story rather than the awkward “yes, the click numbers are small, but trust us” pitch.

Eight GEO Metrics Worth Tracking in 2026

Search Engine Land published a list of eight Generative Engine Optimisation metrics that brands should be tracking by mid-2026: AI citation frequency, citation share by engine, conversational visibility, query coverage, answer share of voice, narrative consistency, conversion influence, and source-quality positioning.

A couple of these were already standard (citation frequency, citation share). The interesting additions are “narrative consistency” (does the AI describe your brand the same way across ChatGPT, Perplexity, Gemini and AI Mode?) and “conversion influence” (can you trace AI visibility to downstream signed-up or purchased outcomes via cohort analysis?). Both are hard to measure cleanly and will reward early adopters.

Most agency client programmes can start with the first three as a baseline and add the others quarterly as tooling catches up. Pure citation counts on their own are too noisy to be trusted as the sole measure of AI visibility.

Google Tests a Gemini 3 Model Picker Inside Search

Google is testing a Gemini 3 model selector inside the search bar itself, with options visible for Pro, Fast and Auto. Sachin Patel and Gagan Ghotra on X both surfaced screenshots; Barry Schwartz reports he hasn’t been able to replicate the test yet.

If this rolls out broadly, it’s a meaningful signal: Google is exposing the model-tier choice (latency vs reasoning depth) directly to the searcher, the same approach Anthropic and OpenAI have already adopted for their consumer apps. The Auto option suggests an automatic router based on query complexity, which is sensible default behaviour.

The practical question for SEO and PPC measurement is whether different Gemini 3 tiers will surface different sources or different ad inventory. We don’t know yet. But set up tracking now that distinguishes AI Mode answers by model variant once the test ships, rather than treating “AI Mode” as a single channel.

Google Ads Launches Journey-Aware Bidding and Demand-Led Pacing

A 27% average increase in unique converting users on Search. That’s the Google-reported headline number for Smart Bidding Exploration, which is now expanding to Performance Max and Shopping campaigns alongside two genuinely new beta features in Google Ads.

Journey-Aware Bidding is the more interesting of the two. It feeds non-biddable conversion signals (newsletter opens, page-level engagement, store visits) into the bidding system so the model has a fuller picture of what eventually leads to revenue, not just the form-fill or transaction itself. According to Google, the goal is to stop the optimiser from over-rewarding low-funnel actions when the actual conversion path is multi-touch.

Demand-Led Budget Pacing is the second beta. It automatically adjusts daily spend based on real-time demand patterns, leaning into peak windows and pulling back in slow ones, while staying inside the campaign’s overall budget. Google cites a 66% reduction in manual budget adjustments among advertisers using total budgets, which is the kind of operational metric that’s easy to report and harder to interpret.

The trade-off is the same one all AI bidding shifts present: less hand-on-tiller, more model trust.

Practical setup for PPC teams: enable Smart Bidding Exploration on Performance Max and Shopping where you’ve previously held it off Search-only, and test Journey-Aware Bidding on accounts where you have well-defined non-biddable conversions in GA4. Trial Demand-Led Pacing on retail accounts with strong promotional cycles, where the manual pacing wins are highest. We’d avoid it on tightly-budgeted lead-gen until the model has a quarter or two of feedback.

AI Max Account-Level Exclusions Arriving as DSA Migration Bites

After the Dynamic Search Ads sunset announced in March, advertisers have been stress-testing AI Max as the replacement and finding the control surface narrower than what DSA offered. Gabriele Benedetti’s complaint, echoed across LinkedIn this week, is that AI Max lacks the granular URL-path and category-based targeting that let DSA campaigns mirror website architecture.

Google Ads Liaison Ginny Marvin acknowledged the gap in a public response. AI Max does support URL rules and combinations, page feeds with custom labels, ad-group-level URL inclusions, and campaign-level URL exclusions. But “page contains” conditions and a few other DSA targeting rules don’t map across yet. Existing DSA URL rules carry over as read-only to migrated accounts, which is functional but not editable.

The compensating announcement is that account-level content and title exclusions are coming to AI Max “later this year.” That’s a meaningful concession because it means brands can lock out content categories (think: clearance, refurbished, regulated SKUs) at the account level rather than fighting them campaign by campaign.

Our take on agency-side migrations: hold off on full DSA-to-AI-Max migrations for accounts that depend heavily on URL-path-based campaign structure until the account-level exclusions land. For accounts where DSA was always loosely scoped, migrate now and use page feeds to enforce structure. The bigger strategic question (whether AI Max’s reduced control is a feature or a bug) probably won’t resolve until late 2026, when we can compare 6-month performance windows on properly migrated accounts.

ChatGPT Ads Manager Officially Goes Self-Serve

The beta self-serve Ads Manager for ChatGPT Ads, which Search Engine Roundtable surfaced in late April before launch, is now officially open to US advertisers. OpenAI confirmed the rollout on 6 May. Pricing follows the CPC model already detailed in last week’s update, with a $50,000 minimum and CPCs in the $3-5 range for early test campaigns.

What’s new in the official launch is the workflow itself. Advertisers can now self-sign up, build campaigns, set budgets, and access conversion tracking inside a familiar dashboard, the kind of thing that was previously a sales-touch process. That’s the moment a new ad surface starts moving from experimental to addressable in normal media planning.

For our part, we’re cautiously running a couple of test campaigns to benchmark ChatGPT ad performance against Google AI Mode placements. Early read: assume nothing, measure everything, and don’t compare CPMs across the two until you’ve controlled for query type. We’ve seen $4 CPCs convert and $3 CPCs not convert on the same product in the same week, which is enough to keep us nervous about anyone making early “ChatGPT is more expensive” claims.

Microsoft Ads Adds Full Conversion Metrics to Custom Columns

Microsoft Advertising expanded its Custom Columns feature to include all conversion metrics, not just the limited subset previously available. That means advertisers can now build account-specific reports combining conversion volume, conversion value, conversion rate, cost-per-conversion, and ROAS variants alongside any other metric in the platform.

It’s the sort of release that’s more useful than it sounds, especially for agencies running standardised reporting templates across accounts. If you’ve been forced to do conversion-metric maths in Excel after exporting Microsoft Ads data, that workaround is now redundant. Set up the custom columns inside the platform once and they’ll flow into scheduled exports.

Combined with the Microsoft Ads Performance Max URL-level reporting added last week, the Microsoft platform’s reporting parity with Google Ads keeps closing.

Call Recording Defaults to Yes Across Google Ads

Are your Google Ads call recording settings explicitly chosen, or sitting on the default? They need to be explicit. Google notified eligible advertisers this week that as of 1 July 2026, the call recording setting will default to “Yes” if no explicit choice has been made. Recording-on means inbound calls from call assets and calls-only ads will be captured for AI-powered call quality scoring, lead qualification, and the AI-Qualified Call Leads feature rolled out in late April.

It’s a good default for accounts that have already aligned with their compliance teams. It’s a poor default for accounts in regulated verticals (finance, healthcare, legal) where consent and recording-disclosure rules vary by jurisdiction. If you manage Google Ads for any client in those categories, audit the call recording setting on every account before 1 July and document the explicit choice in writing.

UCP Checkout Lands in Main Google Search Results

Three months after Google added UCP-powered checkout to AI Mode, the same feature has now expanded to free product listings in main Google Search results. Brodie Clark spotted it first on Wayfair listings, with the expectation that Etsy and Target rollouts will follow.

The user flow is the kind of thing Google’s been moving towards for years. Tap “Buy” inside a free product listing, billing and shipping pre-fill from the user’s Google Pay credentials, transaction completes without leaving the SERP. For retailers integrated with UCP, that’s a measurable conversion-rate uplift on Google-sourced traffic. For retailers not integrated, it’s a competitive disadvantage that scales with category.

For SEO and PPC working together, this is the second time in 18 months that a SERP feature has flipped commercial intent into transaction completion at the search layer. The first was Buy on Google’s earlier iterations. UCP is the shipping-credentials version of the same idea, and this time the pipes (Google Pay penetration, retailer integrations, AI Mode commerce) are mature enough to make it stick.

Google Business Profile Suspensions Spike Over Account Restrictions

A wave of Google Business Profile suspensions hit the local SEO community over the past several days, traced to underlying user-account restrictions rather than profile-content violations. The pattern, surfaced by Barry Schwartz and several local SEO practitioners, is that profiles are being suspended even when the listing itself is fine. The trigger appears to be on the owning Google account, not the place.

Google hasn’t confirmed whether this is a bug or an enforcement action. Local SEO teams report a mix of immediate reinstatements via standard appeals and longer drawn-out cases requiring direct contact with Google support.

If you manage GBP for clients, the practical step this week is to audit the access list on every active profile, verify there’s at least one active manager-level account that’s separately not flagged, and document the chain of access in case you need to recover from a sudden suspension.

AdSense Drops Back-Button Trigger for Vignette Ads

The 15 June 2026 deadline for back-button hijacking compliance now has its first concrete AdSense-specific instruction. Google confirmed this week that vignette ads in AdSense, the full-page interstitials that have historically loaded on the back button, will have that trigger disabled by 15 June. After that date, vignette ads will only load on natural page navigation forward, not on the back action.

If you’re running AdSense and haven’t already audited which placements use vignette ads, this is the week. Sites that depended on back-button vignettes for monetisation will see revenue drop unless the placements are reconfigured to use forward-page or sticky formats. The compliance deadline applies regardless of whether the vignette behaviour is intentional or default.

This is the second concrete touchpoint of the back-button policy from the 13 April spam policy update. Expect more enforcement specificity in the next four to six weeks.

WordPress Slips as Astro's Weekly Downloads Hit 2.5M

WordPress’s market share peaked at 43.6% in mid-2025 and now sits at 42.2% as of this week, per W3Techs data, a 1.4-percentage-point drop. Astro’s weekly downloads have doubled year on year to 2.5 million, up from 1.4 million in the equivalent week of 2025. Cloudflare’s acquisition of Astro earlier this year accelerated adoption among engineering-led teams.

Two things to keep in perspective. First, 1.4 points off the dominant CMS in the world is still a 1.4-point shift on a base of millions of sites, which is meaningful but not collapse. Second, 10.56% of WordPress sites haven’t been updated since 2022, so the active-site share is meaningfully below the headline number.

For agency-side decisions, this matters most when advising on platform choice for a new build or a replatform. WordPress 7.0 with native AI integrations is in development and may slow the bleed, but for high-performance content sites with engineering teams who can support a static-first stack, Astro’s getting hard to ignore. We’d still default to WordPress for marketing-led teams without engineering capacity.

Strategic Direction: Where Search Marketing Is Heading

Three themes pull this week’s announcements together.

The first is that AI search is ceasing to be a layer over traditional search and becoming its own substrate, with its own index, its own attribution rules, and its own legal exposure. Microsoft’s grounding-vs-indexing post says this explicitly. Google’s AI Mode link overhaul says it implicitly through five UI changes that treat citations as first-class objects. The MacIsaac lawsuit will say it through the courts.

The second is that ad platforms are converging on AI-mediated bidding faster than agencies can adopt the new mental models. Journey-Aware Bidding, demand-led pacing, AI Max account-level exclusions, ChatGPT’s self-serve Ads Manager, AI-qualified call leads. Each one of those reduces the surface area where humans set the rules and increases the surface area where humans set the inputs and verify the outputs. That’s a different job description than the one most account managers were hired into.

The third is that publisher economics are hardening into a clear two-tier outcome. Tier one (subscription-labelled, author-attributed, author-credible publishers) get rewarded inside AI Mode with click-likely placements. Tier two (low-quality, scaled, pattern-matched content) gets filtered, deprioritised or, in the worst case, defamation-adjacent. Forrester and Taylor’s pieces this week both argue, from different angles, that the moat is editorial control, not content volume. We agree.

Fourth-quarter 2026 strategic plans for agency clients should reflect those three trajectories whether the immediate KPIs do or not.

Key Takeaways

Audit your sites’ AI Mode citation behaviour weekly, particularly subscription labelling, author attribution, and inline-link selection. Google’s five-way update means small content-level changes affect citation placement.
Treat freshness as an AI-grounding signal, not just a ranking one. Refresh evergreen pages older than 18 months in critical categories to reduce stale-content grounding risk.
On Google Ads accounts with strong non-biddable conversion data in GA4, pilot Journey-Aware Bidding. Trial Demand-Led Pacing on retail with promotional cycles. Hold off on tight-budget lead-gen.
For DSA-to-AI-Max migrations, keep DSA structures in place if they depend on URL-path control. The promised account-level content and title exclusions arriving later in 2026 will make migrations cleaner.
Audit Google Ads call recording settings for clients in regulated verticals before 1 July. The default flips to “Yes” on that date.
For brands publishing on LinkedIn or Medium, the visible author name is now a citation signal in AI Overviews. Personal brand investment matters more for these platforms than domain-level authority.
For any client whose name is shared with a notorious individual, audit AI Overview output regularly. The MacIsaac case puts AI Overview defamation on the legal map.

Frequently Asked Questions

What changed about Google’s AI Mode and AI Overview links this week?

Google rolled out five updates: further-exploration suggestions at the end of answers, subscription labels next to news outlets users already pay for, community perspectives previewed inline, contextual links placed next to relevant text, and desktop link previews on hover. The pattern is that citations are getting treated as first-class UI rather than footnotes.

What does Microsoft mean by “grounding” versus “indexing”?

A traditional search index is built to rank pages so users can pick what to trust. A grounding index is built to support a single AI-generated answer, which means it has to assess factual accuracy, freshness, source clarity, and conflict between sources before generation. Microsoft’s argument is that the two functions need different infrastructure.

Does the MacIsaac lawsuit mean Google can’t show AI Overviews any more?

Not directly. It’s a single defamation claim in Canada, and it’ll take time to resolve. But it’s the first sizeable test of whether AI Overviews carry publisher-level liability for generated claims, separate from the linked sources. If MacIsaac wins or settles favourably, expect Google to tighten the sourcing standard inside AI Overviews quickly.

Should I move my Google Ads campaigns to AI Max now?

Depends on the account structure. If your DSA campaigns rely on URL-path-based targeting that mirrors site architecture, hold off until Google ships the promised account-level content and title exclusions later this year. If your DSA campaigns are loosely scoped, migration to AI Max is fine now and page feeds with custom labels will give you most of the structure you need.

Is WordPress losing relevance, or is this a slow drift?

Slow drift, not collapse. WordPress is down 1.4 percentage points from a mid-2025 peak, and a meaningful chunk of its share is abandoned sites that haven’t been updated since 2022. Astro’s gaining on engineering-led teams. For marketing-led teams without dedicated engineering, WordPress is still the default.

Conclusion

This week’s news cycle has a clear theme. AI search is no longer something agencies can plan around as a single channel. It’s an index, a citation system, a legal liability, an ad surface, and a measurement framework, each evolving on its own timeline. The announcements from 6-7 May (Google’s AI Mode link overhaul, Bing’s grounding manifesto, Journey-Aware Bidding, the MacIsaac lawsuit) all point at the same conclusion: the surface area of search marketing has gotten wider, and the muscle memory of the last decade only partially transfers.

For most agency clients, the practical move this quarter is to add citation-quality measurement to standing reporting, audit AI Overview attribution for any brand with reputational risk, and pilot the new Google Ads bidding features on appropriate accounts. The 21st-century version of search is here, and it has more moving parts than the last one.

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

Table of Contents

SEO & Algorithm Updates

AI Search, GEO & AEO

PPC & Paid Media

Commerce, Local & Compliance

Industry

Wrap-Up