29% of Search Volume Is Declining, AI Overviews Aren’t Filtering Junk Clicks, and ChatGPT Changes Who It Cites
Estimated reading time: 17 minutes
A data-heavy week in search marketing news, and most of it points the same way. A study of a million keywords put hard numbers on how AI is rewiring search demand, a fresh experiment picked apart Google’s favourite defence of AI Overviews, and new research showed ChatGPT citing a completely different set of brands the moment it starts thinking harder. There’s plenty for PPC teams too: Performance Max diagnostics, branded-traffic defence, and signs that image and video ads are heading to ChatGPT. Anthropic shipped new models, and we lost a genuine SEO pioneer. Let’s get into it.
Table of Contents
SEO Updates & Search Data
A million keywords show 29% of search demand in decline
Study: AI Overviews' lost clicks weren't lower quality
Google data: AI search users have moved past keywords
AI Search, GEO & Measurement
ChatGPT's thinking mode changes which brands get cited
Microsoft: rankings and citations are not the same metric
The Open Web Under Pressure
Habitual publisher traffic is collapsing
Google's DMCA crisis is disrupting the web
Google sends searchers straight to publisher-hosted AMP pages
PPC & Paid Media
Performance Max gets Channel Diagnostics
How competitors target your branded traffic
OpenAI hiring points to image and video ads in ChatGPT
AI Tools & Industry
Anthropic ships Claude Sonnet 5 and returns Fable 5
Agents and MCP servers quietly automate SEO
A generational handover: Bruce Clay and Fabrice Canel
Wrap-Up
Strategic Direction: Where Search Marketing Is Heading
A Million Keywords Show 29% of Search Demand in Decline
Start with the study everyone will be quoting for the rest of the year. Fractl and Search Engine Land analysed more than a million high-volume keywords across 379 brands in eight verticals, roughly 35 billion monthly searches, and paired it with a survey of just over a thousand US consumers. The headline: 29% of search volume is now in decline, four points worse than Gartner predicted back in 2024.
But the more useful finding is buried in the detail. This isn’t a simple collapse. Around 40% of tracked keywords lost meaningful volume, while about 20% grew, and because the growing terms carry more searches each, the net change across the set was close to flat. Search demand is redistributing, not disappearing.
Where you sit in the market decides which side of that line you land on. FinTech got hit hardest, down nearly 38%, with health and wellness close behind, because those categories are overwhelmingly non-branded and informational, exactly the queries an AI answer swallows whole. Transactional verticals held up or grew: SaaS search volume was up 48% year on year.
The pattern is consistent: informational and how-to queries are draining toward AI, while branded and commercial intent stays put. If your content strategy leans on top-of-funnel how-to guides, this is the data that should reshape your next planning meeting.
The practical move isn’t to mourn the lost volume. It’s to find the keywords gaining ground inside your own category and build the authority signals, first-party data, named expertise, genuine reviews, that earn a place in both the classic results and the AI answers. Demand is moving. The job is to move your content to where it goes.
Study: AI Overviews' Lost Clicks Weren't Lower Quality
Google has spent months arguing that when AI Overviews cost you clicks, they’re only trimming the low-quality ones people would have bounced from anyway. A new experiment says that’s not true.
Researchers Saharsh Agarwal and Ananya Sen ran a randomised field study, updated on SSRN, and found that AI Overviews cut organic clicks by around 39.8%. Then they checked whether the clicks that survived were any better than the ones lost. They weren’t. Across bounce rate, quick returns to the results page, and time on site, there was no measurable difference between the two groups.
“No measurable difference in bounce rates, search returns, or time on site.”
– Saharsh Agarwal and Ananya Sen, working paper on SSRN
That directly contradicts Google VP Liz Reid, who framed AI Overviews as filtering out low-engagement “bounce clicks.” The data shows the lost clicks were just as engaged as the ones that stayed.
The effect concentrates where you’d expect: AI Overviews triggered on 41% of all queries but 53% of informational ones, with navigational and transactional searches barely touched. And when overviews were removed, the top three organic results picked up the most clicks, which tells you exactly whose traffic AIO is absorbing.
Keep this study handy. Next time someone waves away an AI Overviews traffic drop as “just the junk clicks,” you’ve got a peer-reviewed answer that says otherwise.
Google Data: AI Search Users Have Moved Past Keywords
Tied to that redistribution story is a shift in how people actually search when an AI is answering. Google’s May report shows AI Mode queries running about three times longer than traditional searches, with users typing full conversational prompts instead of clipped keywords.
The example Google gives is telling. Instead of “best running shoes 2026,” someone now types: “I’m training for my first 5K and I’ve never bought running shoes before, which pair should I start with?” That’s not a keyword. It’s a paragraph of context, and it changes what your content needs to answer.
A few numbers stand out:
- Follow-up queries are growing 40% month on month, so people stay in the conversation rather than taking the first answer
- Multimodal searches, by voice, image or video, are now one in six queries
- Planning-style queries are growing 80% faster than AI Mode overall
Most content is still written for the three-word keyword era. The gap between how people search and how we optimise is widening. The fix is to write for the questions people actually ask, map the likely follow-ups, and make sure your images and video are indexed properly, because visual input is the fastest-growing query type there is.
ChatGPT's Thinking Mode Changes Which Brands Get Cited
Here’s the GEO finding of the week, and it’s a big one. When ChatGPT switches into high-reasoning mode, it cites almost entirely different sources. An analysis found just 25.6% of cited domains overlap between minimal and high reasoning. Turn on deeper thinking and roughly three-quarters of the sources change.
Why? Because the model works harder. In high-reasoning mode ChatGPT ran around 1,130 web searches versus 245, and cited 4.5 sources per answer instead of 2.6. It also shifts what it trusts: Reddit’s share of citations dropped from 15% to 7%, while government, academic and official documentation climbed sharply. Deeper reasoning rewards primary sources over forum chatter.
For anyone doing GEO work, this reframes the problem. A complex comparison query, say evaluating CRM platforms, fired off 24 sub-queries in high-reasoning mode against 5.5 in the quick version. Your brand might surface in the fast answer and then vanish once the model starts drilling into specifics, or the other way around.
That’s a genuinely awkward measurement problem. Being cited once tells you very little now, because the citation set depends on how hard the user asked the model to think. The takeaway: get your documentation, spec pages and third-party references in order, so you survive the deep-research passes, not just the surface-level ones.
Microsoft: Rankings and Citations Are Not the Same Metric
Microsoft made a similar point from a different direction, and turned it into a product. Bing Webmaster Tools now has two separate dashboards: one for traditional search rankings, one for AI citations. By splitting them, Microsoft is saying out loud that these are different things.
The distinction matters. A ranking measures how your page compares to competitors for a query. A citation measures whether a specific passage on that page is useful as evidence for an AI answer. As Microsoft put it, what makes a page rank well for a human isn’t the same as what makes a passage useful to an AI.
Under the hood, Bing scores passages on completeness, freshness and authority, and the new Citation Share metric shows your slice of citations for a given grounding query. The catch, as ever with first-party tools, is that it only reflects Bing’s own surfaces, not your visibility across ChatGPT, Gemini and the rest.
Still, the practical advice is sound and applies everywhere: stop treating rank position and citation count as the same number. Track the gap between where you rank and where you get cited, and watch how that delta moves. It’s becoming one of the more honest signals of how AI systems actually use your content.
Habitual Publisher Traffic Is Collapsing
A quieter piece this week named something a lot of publishers are feeling but struggling to measure: habitual traffic is collapsing. That’s the steady, unthinking return visits, the people who used to type a familiar site into the bar or click the same bookmark out of routine, now increasingly satisfied by an AI summary before they ever arrive.
This is the flip side of every zero-click and AI Overviews stat we’ve covered. When the answer comes to the user, the habit of visiting the source erodes, and habits are what publisher business models quietly depend on.
There’s no neat fix, but there is a direction. The sites that hold onto habitual visitors will be the ones that offer something an AI summary can’t reproduce: a distinct voice, original reporting, tools, community, a reason to come back that isn’t just information retrieval. If your entire value proposition is “we have the answer,” the answer engine has already replaced you.
Google's DMCA Crisis Is Disrupting the Web
Something uglier is happening in the plumbing of search. Google is facing a growing wave of fraudulent DMCA takedown requests, fake copyright claims filed specifically to bury legitimate content in the results. Former Googler Pedro Dias put it bluntly, saying Google has a serious problem that no one is working to fix, or willing to.
The mechanics make it nasty. A bad actor files a false claim against a competitor’s page or an unflattering article, and the legally required processing window, often 10 to 14 business days, means the content can vanish from results while the dispute plays out. Recovery means filing a counter-notice and waiting. Suing the filer is usually pointless because they hide behind fake contact details.
For site owners, this moves DMCA abuse from a rare annoyance to a real competitive risk. It pays to know the counter-notice process before you ever need it, and to monitor your own visibility for sudden unexplained drops on specific URLs that could signal a bogus claim rather than an algorithm change. The system was built in 1998 for a different web. Right now it’s being turned against the people it was meant to protect.
Google Sends Searchers Straight to Publisher-Hosted AMP Pages
A smaller technical change, but a symbolic one. From 1 July, Google Search sends users straight to publisher-hosted AMP pages rather than routing them through Google’s cached AMP viewer. Google frames it as reducing the maintenance and analytics headache for publishers, and says rankings are unaffected: AMP content ranks like any other page.
Honestly, this reads like the slow winding-down of AMP’s special status, which has been fading for years anyway. Fewer people build AMP pages now, and the format long ago lost the ranking perks that once justified the hassle.
If you still maintain AMP, the change is good news: simpler analytics, cleaner tracking, one less Google intermediary between you and your own visitors. If you dropped AMP years ago, this is confirmation you didn’t miss much. Either way, it’s a reminder that the technical scaffolding of search keeps quietly shifting under our feet.
Performance Max Gets Channel Diagnostics
Over in paid media, Google added Channel Diagnostics to Performance Max, and it’s a genuinely useful bit of transparency for a campaign type that’s often felt like a black box. The tool flags missing or disapproved assets that are holding back delivery across a channel, so you can see why PMax isn’t spending or serving where you expected.
Anyone running Performance Max knows the frustration: budget sitting idle, or delivery skewed to one surface, with little insight into why. Channel Diagnostics won’t hand you full control, Google still guards that tightly, but it does give you a checklist of fixable problems rather than a shrug.
Practically, make this part of your PMax quality-assurance routine. Check diagnostics when a campaign launches, when spend stalls, or when performance shifts, and clear any flagged asset or approval issues before you start second-guessing bids and budgets. It’s the closest thing to an under-the-hood view PMax has offered in a while, so use it.
How Competitors Target Your Branded Traffic
A practical PPC reminder landed this week on how competitors poach your branded search traffic, and it’s a good prompt to audit your own defences. Rivals use dynamic keyword insertion, comparison pages and brand-modifier terms to intercept people who are literally searching for you, all within Google’s policies.
If you’re not bidding on your own brand, you’re leaving the door open. Brand terms are cheap, they convert, and an undefended brand SERP is an invitation for a competitor to sit above your organic listing with a “why pay more?” message aimed straight at your customers.
The counter-play is straightforward: run a branded campaign, monitor the auction insights report for competitors appearing on your terms, and consider comparison content of your own that pre-empts the “you versus them” queries. Brand defence isn’t glamorous spend, but losing a ready-to-buy customer at the final step because a rival outbid you on your own name is the kind of leak that quietly adds up.
OpenAI Hiring Points to Image and Video Ads in ChatGPT
The signals keep pointing to ChatGPT becoming a serious ad platform. OpenAI’s recent hiring, particularly for ads and media roles, suggests image and video ad formats are coming to ChatGPT, including native and conversational placements.
We flagged the arrival of product-feed ads a few weeks ago. Visual and video formats would be the logical next step, turning ChatGPT from a text-answer box into something that looks a lot more like a full advertising surface, sitting inside the assistant millions now use daily.
Nothing has shipped yet, so treat this as a planning signal rather than a to-do. But if you run visual campaigns, it’s smart to start thinking now about how your creative might work inside a conversational assistant, where an ad appears mid-answer rather than beside a set of results. The teams that experiment early on new surfaces usually get the cheap attention before everyone else piles in.
Anthropic Ships Claude Sonnet 5 and Returns Fable 5
Anthropic had a busy week on the model front. It rolled out Claude Sonnet 5, described as offering near-Opus intelligence across all plans with introductory pricing, and brought back Claude Fable 5 with new usage limits and additional safety classifiers after its earlier debut.
Why does a model release belong in a search roundup? Because these are the engines increasingly doing the work: drafting content, analysing data, running the agents that are creeping into SEO and PPC workflows. A cheaper, more capable model tier changes the economics of what you can automate, and the pace of releases, several in a matter of weeks, is itself the story.
For most marketing teams the practical read is simple. The cost of capable AI assistance keeps falling while quality keeps rising, which widens what’s realistic to hand to a model. If you tested an AI workflow six months ago and found it not quite good enough, the tier available today is a different proposition. It’s a good moment to retest the things you shelved.
Agents and MCP Servers Quietly Automate SEO
Automation was the quiet throughline across the SEO tool blogs this week, and it’s becoming the real story rather than a novelty. Ahrefs published a walkthrough of how its own team taught an AI agent to keep data-driven content fresh, automating the monthly slog of updating statistics and figures inside existing articles. That’s a genuinely tedious job handed to a machine that does it well.
Alongside that, SE Ranking put out a guide to the best SEO MCP servers, the connectors that let AI assistants plug directly into SEO data and tools. When your assistant can query live ranking data or crawl a site through a standard interface, the gap between “asking about SEO” and “doing SEO” starts to close.
Put the two together and the direction is obvious: the routine, repetitive parts of SEO are quietly being automated, and the practitioners who learn to build and supervise these agent workflows will get a lot more done than those doing it all by hand. This is the least dramatic trend in this week’s news and possibly the most consequential for how the job actually changes.
A Generational Handover: Bruce Clay and Fabrice Canel
Finally, two people who shaped this industry stepped out of it this week, one sadly and one happily. Bruce Clay, one of the true pioneers of SEO, passed away, and the tributes across the community were a reminder of how much of what we take for granted, the very idea of optimising for search as a discipline, traces back to a small number of early practitioners like him.
Separately, Fabrice Canel, a long-time search leader at Microsoft and the driving force behind IndexNow, announced his retirement. Between them, the two departures mark a generational handover: the people who built the foundations of search stepping back just as AI rewrites the surface they helped create.
It’s a fitting moment to take stock. The fundamentals these figures championed, crawlable sites, clean structure, genuine relevance, quietly still underpin everything, even as the interface on top of them changes beyond recognition. A fitting moment, too, to remember the people who wrote the first rules.
Strategic Direction: Where Search Marketing Is Heading
Pull this week together and the theme is measurement catching up with reality. For two years the industry has talked about AI eroding search traffic; this week it got the receipts. A million keywords showing demand redistributing, a field study dismantling the “junk clicks” defence, Microsoft splitting rankings from citations in its own tools, ChatGPT changing its sources the moment it thinks harder. The debate is moving from whether AI is remaking search to precisely how much, and where.
The strategic response splits cleanly in two. First, measurement has to evolve: rank tracking alone is now a partial picture, and the teams pulling ahead are the ones tracking citations, AI-referral traffic and the gap between ranking and being quoted. Second, content has to earn its place in AI answers, which rewards primary sources, first-party data and genuine expertise over the thin how-to content that AI reproduces for free.
Underneath both sits a harder question about defensibility. Habitual traffic is fading, informational queries are draining to AI, and the sites that survive will be the ones offering something an answer engine can’t: original data, a real voice, tools, trust. For the back half of 2026, the planning question isn’t “how do we rank” so much as “why would anyone, human or model, choose us as the source.” Answer that well and the rankings tend to follow.
Key Takeaways
- Search demand is redistributing, not vanishing: 29% of volume is declining while commercial and branded queries hold up. Audit which of your category’s keywords are growing and shift content toward them.
- A new field study found AI Overviews cut organic clicks by about 40%, and those lost clicks were just as engaged as the ones kept. Google’s “we only trim junk clicks” defence doesn’t hold.
- ChatGPT’s high-reasoning mode changes roughly 75% of its cited sources and favours primary and official documentation over forums. Get your spec pages and third-party references in order.
- Rankings and AI citations are different metrics. Bing Webmaster Tools now separates them; track the gap between where you rank and where you get cited.
- AI Mode queries are three times longer and conversational. Rewrite key content around the full questions people ask, and index your images and video for multimodal search.
- PPC: use the new Performance Max Channel Diagnostics to fix delivery blockers, and defend your branded search terms before a competitor bids on them.
- Automation is arriving in SEO for real, through content-refresh agents and MCP servers. Start learning to build and supervise these workflows.
Frequently Asked Questions
Is search traffic actually declining because of AI?
It’s redistributing more than declining. A million-keyword study found 29% of search volume falling, concentrated in informational and non-branded queries, while transactional and branded terms held steady or grew. The net change across the set was close to flat, so the task is to follow demand to the keywords and formats that are gaining.
Do AI Overviews only remove low-quality clicks, as Google suggests?
A recent field study says no. It found AI Overviews cut organic clicks by around 40%, with no measurable difference in engagement between the clicks lost and those kept. The lost traffic was just as valuable, which undercuts the argument that AI Overviews simply filter out bounce clicks.
Why does ChatGPT cite different sources in reasoning mode?
Because it searches far more and evaluates more sources, shifting toward primary and official documentation. Only about a quarter of cited domains overlap between quick and deep-reasoning answers, so a brand can appear in one and not the other. Consistent documentation and third-party references help you survive the deeper passes.
How is DMCA being abused against websites?
Bad actors file fake copyright takedown notices to push legitimate content out of Google’s results. Because the process has a built-in delay and filers often hide behind false contact details, targeted pages can disappear for one to two weeks. Knowing the counter-notice process in advance is the best defence.
Should marketers care about new Claude model releases?
Yes, indirectly. Cheaper, more capable models lower the cost of automating content, analysis and agent-driven SEO and PPC tasks. If an AI workflow fell short for you six months ago, the current model tier may now clear the bar, so it’s a sensible moment to retest.
Conclusion
If one idea ties this week together, it’s that the industry finally has hard numbers for what it’s been sensing. AI isn’t quietly nudging search; it’s redrawing where demand goes, whose clicks survive, and which sources get quoted in the answers people now read instead of visiting. None of that means search is over. It means the definition of winning is changing, from ranking first to being the source both people and models trust.
The teams who thrive through the rest of 2026 will measure citations as seriously as rankings, write content that earns its place in AI answers, and give visitors a reason to choose them that no summary can copy. The data has caught up with the story. Now it’s on us to act on it.
Need help adapting your search strategy for the AI era? Contact the Anicca team for expert SEO and PPC guidance.
