Search Console Socials, ChatGPT Referrals +36.7%

Estimated reading time: 27 minutes

Google spent the week reminding everyone it still owns the front door of the internet, then quietly rebuilt several rooms behind it. This week’s search marketing news covers Search Console’s new social platform tracking, a record-breaking moment for Google Search during the World Cup, an all-time peak in ChatGPT referral traffic, and a Smart Bidding change that PPC teams need to act on before mid-August. There’s also a stack of new research on how AI assistants pick, cite and drop brands. Let’s get into it.

Search Console Now Reports on Instagram, TikTok, X and YouTube Content

Google has launched platform properties, a new Search Console property type that reports on how your Instagram, TikTok, X and YouTube posts perform in Google Search and Discover. Once a social account is verified, its data appears in the Performance report (clicks, impressions, queries, exportable like any other property), the Insights report, and the achievements section.

Verification takes minutes:

Open the property selector in Search Console and click “Add property”
Choose one of the four platforms: Instagram, TikTok, X or YouTube
Follow the on-screen steps to authorise the connection

The rollout is gradual over the coming weeks, so don’t panic if the option hasn’t appeared in your account yet. Barry Schwartz noted that his X profile showed up without manual verification, which suggests Google is pre-linking some known accounts. Google ran a lighter version of this in 2025 called social channels in Search Console Insights; this goes much further by pushing social data into the main reporting surfaces.

For agencies, this closes a measurement gap that third-party tools never could. You can finally answer, with first-party data, which queries surface a client’s TikTok rather than their website, and whether that social visibility is growing. It also tells you something about Google’s direction: social and video content is being treated as a first-class search asset, not an afterthought. If your search strategy still stops at the domain boundary, this is your nudge to widen it.

Google Search Just Had Its Busiest Moment in History

For an industry that writes Google’s obituary most months, this one lands awkwardly. Nick Fox, Google’s Senior Vice President of Knowledge & Information, said Search recorded its highest usage ever in the moments after Argentina’s winning goal in Tuesday’s World Cup quarter-final.

“Google Search broke all prior usage records and saw its highest usage in history right after Argentina scored their winning goal in yesterday’s match!”
– Nick Fox, SVP Knowledge & Information, Google

The wider numbers back him up. Statcounter puts Google’s worldwide search share at 91.25% in June, up from 89.5% a year earlier, and above 96% on mobile. Whatever share of informational queries chatbots are absorbing, real-time moments still send people to the search box in record volumes.

There’s a budgeting lesson in that. Plenty of 2026 planning conversations have started from the assumption that Google demand is evaporating. It isn’t. The query mix is shifting, and AI surfaces are reshuffling who gets the click, but the audience hasn’t left the building. Keep the classic SEO and PPC engines running while you build for the new surfaces.

Google Tests goto Tracking URLs on Search Results

Google is testing google.com/goto passthrough URLs on organic search results. Click a result and, instead of going straight to the site, you pass through a Google redirect that forwards you on to the destination. Brodie Clark flagged the pattern, Alex Greenland spotted it as far back as 23 June, and Barry Schwartz has been collecting reader reports for a week, though he can’t replicate it himself yet.

The redirect wraps the destination in what looks like base64-style encoding, which has fed speculation that it’s aimed at the scrapers and AI bots that harvest Google’s results. Whatever the motive, the practical risk sits with measurement. Passthrough URLs can distort referral attribution, and third-party rank trackers that parse result URLs will need to adapt. One scraper operator commented that they simply discard affected SERPs and fetch new ones, which hints at how patchy the test still is.

Nothing to change yet. But if odd google.com referral paths start showing up in your analytics this month, this is probably why, and it’s a story to keep half an eye on.

Merchant Listing Schema Gains Product Category and Sale Duration

A quiet documentation update with money attached: merchant listing structured data now supports the Product.category property and a new sale duration mechanism. Retailers can mark up which category a product belongs to and how long a sale price runs, all in a machine-readable format Google explicitly supports.

Small change, easy win. Category markup helps Google classify products more accurately across Shopping surfaces, which matters as product results spread into AI-flavoured experiences, and sale duration gives time-limited pricing a defined shelf life rather than leaving Google to guess. If you manage ecommerce clients, add both to the structured data backlog and fold them into your product feed review. The retailers who keep their markup ahead of the documentation tend to be the ones who pick up the new rich result treatments first.

Business Profile Appeals Now Take Evidence Uploads

Google has rebuilt the appeals flow for suspended Business Profiles so evidence uploads now happen inside the appeal workflow itself. The old process made you file the appeal first, then upload supporting documents through a separate form, with only 60 minutes to do it. Miss the window and your appeal went in without evidence.

Anyone who has handled a client suspension knows exactly how stressful that hour was. Utility bills, signage photos, business registration documents, all scrambled together against a countdown timer.

Local SEO teams should update their internal playbooks to match: gather the evidence pack before you hit submit, then attach it as part of the appeal. It’s a quality-of-life fix rather than a policy change, but suspensions remain one of the most damaging things that can happen to a local business, so anything that reduces fumbled appeals is welcome.

Mueller to SEOs: Fix Your HTML Before You Mirror It in Markdown

Should you be publishing markdown copies of your pages for AI agents? Google’s John Mueller thinks the trend is solving the wrong problem. Responding to a Bluesky post about sites shipping text versions for LLMs while ignoring heading structure and landmarks for screen readers, Mueller argued that a properly built website already works for AI agents, search engines, LLMs and, above all, people. Bolting on a separate “agent-friendly” version just builds technical debt you’ll have to redo.

The markdown-mirror tactic took off after Cloudflare announced in February that its infrastructure could auto-generate markdown files, positioning them as a token-saving convenience for LLMs. But OpenAI’s own documentation describes OAI-SearchBot as a website crawler and never once recommends markdown. The crawler everyone is trying to please is asking for good HTML.

Roger Montti’s framing is the memorable bit: SEO has always chased visible tactics without checking whether they move anything, from PageRank-7 directory links onwards. Markdown mirrors look like the latest entry in that tradition. Spend the sprint on accessibility and semantic structure instead; every consumer of your site benefits, humans included.

Search Central Live Deep Dive Heads to Barcelona

Google has confirmed Barcelona as the host city for Search Central Live Deep Dive Europe 2026, running from 30 September to 2 October. The location was chosen from community feedback after Google asked where the European edition should land.

The Deep Dive format is the three-day version of Search Central Live, with more technical depth than the usual one-day events and direct access to Google’s Search Relations team. For UK SEOs, it’s a short flight and, given how much of this week’s news involves Google changing measurement and delivery mechanics with limited documentation, the chance to put questions to the people who build these systems has rarely been more useful. Registration details are on the Search Central blog; previous Deep Dives have filled quickly.

ChatGPT Referral Traffic Hits an All-Time Peak

A 36.7% jump in a single month. SE Ranking analysed traffic data from 101,574 websites across 250 countries and territories and found ChatGPT’s share of referral traffic rose from 0.23% in April to 0.32% in May, its highest level on record and 8% above the previous peak from October 2025.

That’s not a trend line. That’s a step change.

The rise appeared in every major region tracked, and the UK did better than most: EU referrals grew 42.7%, the UK 38.7% and the US 23.2%. Timing points to a product-level cause. On 5 May, GPT-5.5 Instant became the default model, and OpenAI launched its beta self-serve Ads Manager around the same time. Several independent observers saw ChatGPT start surfacing more prominent brand links in answers from early May: Similarweb measured referral visits up roughly 150% comparing the week before 7 May with the period after, and Josh Blyskal reported a rise of around 60% to monitored brand homepages. Context makes the acceleration starker; from January to April, ChatGPT’s worldwide referral share grew only about 1.5% in total.

Two practical notes. First, if your GA4 setup doesn’t already segment AI referrals separately, this is the month that stops being optional. Second, ChatGPT’s share of all AI referral traffic actually fell from 79.74% to 75.96% year on year, because Gemini, Perplexity, Copilot and Claude are growing faster still. Track ChatGPT on its own, but not in isolation.

ChatGPT's Hidden Search Pipelines Quietly Swap Your Citations

Two researchers digging through raw ChatGPT search traffic have found internal source-selection labels, named Labrador, Bright, Oxylabs and SERP, sitting behind the answers users see. Which pipeline handles a query turns out to matter a great deal for who gets cited.

Chris Green ran 1,000 prompts up to ten times each, capturing 9,946 completed search runs. Labrador handled 88.1% of primary retrieval, with Bright at 9.9%, Oxylabs at 1.7% and SERP at 0.3%. Most prompts stayed on one pipeline, but 11.6% switched between runs, and when they switched, the citations moved with them: URL overlap between runs dropped from 0.273 to 0.149, roughly 45% lower, with domain overlap down a similar 42%. Suganthan Mohanadasan’s separate inspection of raw traffic confirmed the same labels.

The uncomfortable conclusion for anyone buying or selling AI visibility tracking: the answer you see doesn’t tell you which retrieval system produced it, and the same question on a different day can draw from a different pool of sources entirely. Single-run snapshots aren’t measurement, they’re anecdotes. If you report AI citations to clients, sample each prompt repeatedly and present ranges, and treat sudden citation “losses” with suspicion before anyone rewrites a page over what may just be pipeline churn.

62% of AI Brand Recommendations Vanish After One Follow-Up Question

This story starts with a correction. When Greg Jarboe queried a suspiciously small number in Clovion AI’s new “Surviving the AI Funnel” study, the firm’s COO didn’t defend it; he fixed it. A designer had dropped a zero in layout, turning 330 verified brand-fact contradictions into 33 and 2,040 brands into 204. I’ll be honest: a research team volunteering that kind of correction before publication makes me trust the data more, not less.

The study itself ran 69,120 multi-turn conversations across Claude, ChatGPT and Gemini in 36 B2B software and fintech categories. Re-asking the same opening question (“best CRM tools?”) kept 90% of the recommended brands intact. Adding one ordinary buyer detail, something as plain as “for a small team”, kept only 28%. In other words, 62% of the brands in the first answer were gone by the second. And no, the phrase wasn’t cherry-picked; “for a large enterprise” produced almost identical churn.

The corrected contradiction data adds a useful wrinkle. Claude underclaims a brand’s own features 160 times against 10 overclaims, ChatGPT underclaims 70 times and never overclaims, while Gemini overclaims 80 times against 30 underclaims. Clovion’s working theory is that Gemini leans on marketing material and video, while Claude and ChatGPT lean on documentation and product pages.

The practical reading: being named in an AI answer is not the same as being trusted by it. Your documentation decides whether assistants undersell you, and your marketing decides whether Gemini oversells you. Make both state precisely who your product is for, because that’s the question that empties the list.

AI Overviews Push Deeper Into Commercial Searches

Semrush tracked more than 600,000 US keywords across 10 industries between November and April and found the share of commercial-intent SERPs carrying an AI Overview grew 71% over the six months. Transactional queries moved the other way, down 5%. Google, it seems, is happy to put AI answers on research-mode searches but still keeps them off the final “buy now” click.

The detail that will make PPC teams sit up: SERPs showing both Google Ads and an AI Overview now appear roughly twice as often as a year ago, and keywords with an AI Overview carried a higher average CPC than those without across most industries. Finance saw the sharpest growth of any sector, with commercial queries triggering AI Overviews up 231% in six months.

The evaluation stage of the funnel is exactly where comparison and “best of” content lives, so this squeezes organic visibility at the moment buyers are choosing between options. Pair that with the CPC finding and the message for advertisers is uncomfortable: the auctions getting AI Overview treatment are the expensive ones. Audit your highest-CPC commercial terms and check what the SERP actually looks like now, not what it looked like when the campaign was built.

Ahrefs Tested Self-Promotional Content in AI Search. It Half Worked

Since February, Ahrefs’ Tim Soulo has been running a live experiment: 34 self-promotional pages (listicles, reviews, opinion pieces) published across five domains, promoting a brand-new conference (Ahrefs Evolve) and an established product (Brand Radar), tracked through 9,886 answers from ChatGPT, Gemini, Perplexity and Copilot.

For the unknown brand, it worked. Evolve broke into 72 prompt-and-engine slots where it had never been mentioned, and 82% of those new mentions appeared in answers citing one of the experiment’s pages. Copilot warmed to it fastest, naming Evolve in 39% of relevant answers in March, 57% in April and 65% in May.

ChatGPT never named it once.

For the established brand, the pages barely registered: only 6% of Ahrefs’ new mentions cited them, with the other 94% coming from third-party content the assistants already trusted. And there’s a sting for anyone publishing “best of” listicles about themselves: 43% of answers that cited a conference-promoting page went on to recommend a competing event from the same list. The pattern is blunt: cited is not recommended. Self-promotional content earns its keep when AI has a genuine gap in its understanding of your category. Once you’re known, third-party corroboration does the heavy lifting, which is an argument for digital PR over another self-published listicle.

Primary Research Earns 3.3x More AI Citations

A useful companion to the Ahrefs experiment, from Gauge’s citation dataset: of 301 live pages cited by AI systems across 316 prompts in seven verticals, only 8 qualified as primary research, meaning the original data and methodology actually live on the page. That’s 2.7% of the pages earning 8.4% of the 1,075 citations, or 3.3 times more citations per page than everything else.

The sharper finding is about format. AI doesn’t reward original data generically; it rewards one shape almost to the exclusion of the rest: the benchmark study that answers “which is best”. Comparative numbers, clear methodology, a defensible winner.

For content teams, that reorders the roadmap. One rigorous benchmark built from data you actually own will likely out-earn a year of commentary posts in AI citations, and it compounds, because assistants keep reaching for the page that settles the comparison question. If you have proprietary data sitting in a client’s platform, the brief writes itself.

GPT-Live Puts Web Search Inside ChatGPT Voice

OpenAI has started rolling out GPT-Live, a new generation of voice models for ChatGPT. GPT-Live-1 becomes the default for Go, Plus and Pro users, with a mini version for the free tier. The models are full-duplex, meaning they listen and speak at the same time, interrupt less, and wait when you pause.

The search marketing angle sits in the handoff. Mid-conversation, GPT-Live can pass a question to GPT-5.5 to run a web search, then fold the answer back into the spoken reply, with visual cards on screen for things like weather, stocks and sports. OpenAI says more than 150 million people already talk to ChatGPT every week, so this isn’t a niche interface; it’s a growing front end to web search that produces answers without a visit to the source site.

The open question, as Matt Southern points out, is citations. Text answers show source links beside the response. Whether a spoken answer names its sources, displays them on a card, or leaves them out entirely will decide whether voice search inside ChatGPT can send publishers any traffic at all. Watch that detail as the rollout spreads.

Google Research Targets AI Video Spam at the Account Level

New Google research describes catching AI-generated video spam by clustering accounts and analysing behaviour patterns across them, rather than judging each video on its own. A single AI clip can look perfectly acceptable; a network of accounts publishing on the same schedule, with the same production fingerprints, cannot.

Semrush’s read is that this signals where detection is heading across Google surfaces generally: away from asset-by-asset review and towards pattern-level analysis of who publishes what, how often and how uniformly. For anyone scaling AI-assisted content production, the implication is worth taking seriously. Passing a quality check per piece isn’t the bar any more. Uniformity itself is the fingerprint, so varied formats, varied cadence and genuine editorial input across a portfolio matter as much as the quality of any single asset.

Smart Bidding Will Stop Overdelivering on Your Targets

After two weeks of advertiser pushback, Google has explained what its Smart Bidding change actually means. From 17 August, budget-limited campaigns using Target CPA or Target ROAS will optimise towards the target you set rather than beating it. Today, those campaigns often overdeliver (a \’a350 target quietly converting at \’a335) because the system cherry-picks the safest auctions. Google says that behaviour was never intended.

The follow-up clarifications matter as much as the change:

Google won’t change your Target CPA or Target ROAS settings automatically
Advertisers who want to keep current performance may need to lower targets before the rollout
Account notifications and a new Bid Target Adjustment Tool will identify affected campaigns

The industry debate has been lively. Kirk Williams asked whether Google is building the system to “choose to be dumber” when budget-limited. Mike Ryan’s counter is that Smart Bidding hasn’t been smart so much as conservative, favouring safe auctions over exploring everything that still satisfies the target; the update aligns delivery with what advertisers actually asked for. Aaron Levy focused on scaling: right now, doubling the budget on a campaign converting at \’a38 against a \’a312 target can spike CPA to \’a316 without warning. Predictable beats efficient when you’re trying to grow an account.

Whichever side you take, the action is the same. Audit every budget-limited Target CPA and ROAS campaign in the next few weeks and make a deliberate choice: bank the current efficiency by lowering targets, or accept delivery at the target you set. Doing nothing is also a choice, just not one you want to explain to a client in September.

ChatGPT Ads Adds Audience Targeting

OpenAI is rolling out Audiences as a targeting option in ChatGPT Ads. There’s no help documentation yet, but Barry Schwartz expects it to work like uploaded customer lists, matching emails or phone numbers in the style of Google’s Customer Match.

Put this alongside the self-serve Ads Manager beta from May and a familiar shape emerges: OpenAI is assembling a conventional ad platform piece by piece, and doing it quickly. For paid media teams, the sensible move is to get access early and run small tests while competition in the auction is thin. Nobody is suggesting moving budget wholesale, but the accounts that understand ChatGPT’s ad mechanics before their competitors will be glad they started with pocket-money spends.

Google Answers the Big AI Max Questions

Google has added a set of frequently asked questions to its AI Max help documentation, covering what AI Max is, why to use it, how it differs from Performance Max and what happens when you upgrade a campaign.

The clarification most teams need in writing: AI Max is not a new campaign type. It’s a feature set layered onto existing Search campaigns, which is why upgrading doesn’t reset learning or restructure the account. That distinction has caused genuine confusion in client conversations since the AI Max rollout began, especially with stakeholders who lived through the PMax migration and assumed this was the sequel. If you’ve been fielding “is this Performance Max again?” questions, the FAQ finally gives you an official document to circulate rather than your own diagram drawn on a call.

Travel Campaigns Beta Expands to Things to Do and Events

Google Ads is widening its Search campaigns for Travel beta beyond hotels and flights to cover Things to Do and Events, giving eligible advertisers a purpose-built campaign type for attractions, tours, experiences and ticket sales.

Travel-specific campaign types have historically come with inventory and formats standard Search campaigns can’t reach, so this is more than a rename. If you run marketing for venues, attractions, festivals or tour operators, ask your Google rep about eligibility now rather than waiting for general availability. Betas reward the advertisers who feed them data early: you shape the benchmarks, you learn the quirks, and you’re already optimised when competitors arrive at launch. The summer events season is exactly the wrong time to be on the outside of this one.

92% of B2B Buyers Say AI Shapes Their Vendor Shortlist

How much of the B2B buying process has actually moved into AI tools? Semrush surveyed 643 US business professionals (622 valid responses, with findings based on the 519 who use AI at work) and the answer is: more than most vendor marketing assumes. 84% use AI for work, 69% of those daily, and 66% regularly use it specifically to research products, vendors and solutions.

The headline for anyone selling B2B: 92% say AI has shaped their vendor shortlist, and 45% say it did so significantly. More striking is what doesn’t move the needle. Only 7% notice a vendor in an AI response because they recognise the name. What makes a vendor stand out is how precisely it matches the buyer’s stated use case, which levels the field for smaller and newer players against household names.

Trust runs high but not blind: 75% trust AI vendor recommendations, yet nearly all verify before committing, and 89% expect to rely on AI more for work decisions in future. It’s a US sample, so apply the usual caution before quoting it for UK clients, but the direction of travel is unmistakable, and it connects directly to Clovion’s churn findings above: the shortlist is being written by machines that care about use-case fit, so publish content that states yours in plain terms.

Yoast SEO Now Works With Elementor's Atomic Editor

A small one for the WordPress crowd. Yoast SEO can now read content built in Elementor’s new atomic editor. Until this release, pages built with the atomic editor came back blank in Yoast’s analysis: no text length, no keyphrase checks, no readability score, even on pages full of content.

If you or your clients switched to the atomic editor recently and wondered why every page suddenly failed its SEO checks, that was the reason, and the fix is simply updating the plugin. It’s also a quiet reminder that page builder migrations can break more than layouts; when a builder changes how it stores content, anything that parses that content (SEO plugins, translation tools, search indexing plugins) needs to catch up. Worth a quick audit after any builder change, not just this one.

Strategic Direction: Where Search Marketing Is Heading

The through-line this week is measurement catching up with reality. Search Console now reports on social platforms because discovery genuinely happens there; GA4 setups need AI referral segments because ChatGPT now sends measurable traffic; and the research from Green, Clovion and Ahrefs all says the same thing about AI answers: they’re volatile surfaces, not stable rankings. The winners in AI visibility will be the teams who treat it statistically, sampling repeatedly, reporting ranges, and resisting the urge to react to any single snapshot.

Meanwhile, Google is playing both defence and offence. A record usage moment and 91% share say the core business is healthier than the obituaries claim, while goto redirect tests and account-level spam detection show it hardening the platform against the scraping and synthetic content that AI has industrialised. The paid side tells a parallel story: Smart Bidding is being retuned for predictability over peak efficiency, AI Overviews are spreading across the commercial queries advertisers pay most for, and OpenAI is building an ad platform at speed.

For agencies and in-house teams, the portfolio answer holds: keep compounding the classic SEO and PPC work that still serves 91% of the market, while building the muscles the next surface demands, which increasingly means benchmark-grade original data, precise machine-readable product information, and content that states exactly who it serves.

Key Takeaways

Verify your social profiles as platform properties in Search Console as the rollout reaches your account; it’s the first first-party view of how social content performs in Google Search
Audit budget-limited Target CPA and Target ROAS campaigns before 17 August, and decide deliberately whether to lower targets (keep efficiency) or hold them (gain predictability)
Treat any single AI visibility snapshot as one sample, not the truth: ChatGPT’s retrieval pipeline switches alone can cut citation overlap by around 45%, and one buyer follow-up question drops 62% of recommended brands
Segment AI referrals separately in analytics now; ChatGPT referral share jumped 36.7% in a month, with UK growth at 38.7%
Prioritise benchmark-style original research in content plans; primary data earns 3.3x more AI citations per page, and “which is best” benchmarks dominate
Review your highest-CPC commercial keywords: AI Overviews on commercial-intent SERPs grew 71% and now co-exist with ads twice as often as a year ago
Don’t let AI hype starve core budgets; Google just recorded its highest usage in history and holds 91.25% search share

Frequently Asked Questions

Do I need to own a website to use Search Console’s new platform properties?

No. Platform properties let creators and brands verify Instagram, TikTok, X and YouTube accounts directly, even without a website. Once verified, search performance data for that social content appears in the standard Performance and Insights reports.

Will the Smart Bidding update increase my CPAs?

Possibly, if your campaigns are budget-limited and currently beating their targets. From 17 August, Smart Bidding will optimise towards your set Target CPA or ROAS rather than outperforming it. If you want to preserve today’s efficiency, lower your targets before the rollout; Google’s Bid Target Adjustment Tool will flag affected campaigns.

Why did ChatGPT referral traffic spike in May?

The timing points to product changes around 5 May, when GPT-5.5 Instant became the default model and ChatGPT began showing more prominent brand links in answers. SE Ranking’s data across 101,574 websites shows the rise appeared in every major region, so it wasn’t an isolated fluke.

Do “best of” listicles get my brand recommended by AI assistants?

They get you cited, which isn’t the same thing. In Ahrefs’ experiment, 43% of answers that cited a self-promotional page recommended a competitor from the same list. The tactic works best for new brands that AI doesn’t yet associate with a category; established brands gain far more from third-party coverage.

Should we publish markdown versions of our pages for AI crawlers?

Google’s John Mueller advises against it, and OpenAI’s search crawler documentation asks for normal website access, not markdown. A well-structured, accessible HTML page serves AI agents, search engines and human visitors at the same time, without the maintenance burden of a parallel version.

Conclusion

This week rewarded anyone paying attention to the machinery behind the answers. Search Console opened a window onto social search performance, researchers exposed how arbitrary AI citations can be, and Google told PPC teams precisely how bidding behaviour changes in August. None of it demands panic; all of it demands calendar entries. Verify your platform properties, book the Smart Bidding audit, segment your AI referrals, and put a benchmark study on the content roadmap. The teams that convert news into scheduled work, week after week, are the ones whose clients never notice the ground shifting beneath them.

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

Table of Contents

SEO & Google Search Updates

AI Search & Visibility

PPC & Paid Media

Studies & Tools

Wrap-Up