ChatGPT Linked to 9% Drop in Traditional Search Usage

A new research paper quantifies what many SEOs have suspected: wider access to ChatGPT is associated with a 9% drop in traditional search usage. The study compares how frequently ChatGPT versus Google drives users to websites, providing the first hard numbers on AI's cannibalization of organic search traffic, according to Search Engine Journal.

The trust gap between AI search and traditional search remains significant. A survey cited by Search Engine Journal shows only 28% of Americans trust AI search results, suggesting that while AI tools are pulling users away from Google, a substantial portion of searchers still prefer traditional results — particularly for high-stakes queries. Separately, YouGov's 2026 *Searching for Answers* report (covered by Search Engine Land) found that 86% of Americans still use search engines, while 25% have used AI assistants in the past 30 days, painting a picture of coexistence rather than replacement — for now.

Adding to the opacity, Google's Nick Fox publicly claimed AI Search features drive billions of clicks to websites weekly, but as Search Engine Journal noted, no supporting data was provided and the figure cannot be independently verified. The unverifiable nature of the claim has prompted skepticism across the SEO community.

Following last week's launch of Google Search Console platform properties for Instagram, TikTok, X, and YouTube — covered in our July 6th issue as a straightforward expansion of GSC's scope — Search Engine Journal has now published a critical counter-reading of the feature. The piece argues the social tracking rollout effectively obscures the click loss caused by AI-generated search answers by folding social impressions into reporting and inflating apparent 'visibility' metrics while organic web traffic quietly declines. Beyond the transparency concern, the article raises a data sovereignty issue: verified social profiles effectively hand Google free training data for its AI models. The Ahrefs Digest #281 also flagged the feature, framing it neutrally — a contrast to SEJ's sharply critical take.

New Invoca data, covered by Search Engine Journal, provides the first breakdown of AI referral call performance by channel. Calls originating from ChatGPT convert to leads more frequently than any other channel — though overall conversion rates remain average. The finding suggests that when ChatGPT does send traffic, it tends to send higher-intent users, a meaningful signal for brands evaluating the ROI of AI search visibility efforts.

In a significant confirmation from Google engineers, John Mueller and Martin Splitt stated on the *Search Off the Record* podcast that low-quality AI-generated content can directly cause pages to receive the 'Crawled – Currently Not Indexed' status in Google Search Console. The engineers explicitly linked AI content quality to indexing failures — not just thin or duplicate content — marking one of the clearest on-record statements from Google connecting AI content to indexing problems.

Marie Haynes corroborated the trend from the practitioner side, reporting a notable uptick in sites showing elevated 'Crawled – Currently Not Indexed' counts. While quality is typically the culprit in her observations, she also documents a case where a technical issue (not quality) was the likely cause — a useful reminder that not every occurrence is content-related. Her post links to diagnostic tools for working through the report.

Search Engine Journal's analysis of Mueller's guidance adds further nuance: Google's site quality assessment extends beyond text content to the full user experience — meaning non-indexed pages on a site can still drag down quality signals for indexed pages, even if they never appear in search results themselves. Mass-scale AI content strategies compound these risks by exhausting crawl budgets and failing indexing thresholds in ways teams often don't detect until rankings have already collapsed, per a separate Search Engine Journal deep-dive on 'crawl economics.'

Google's John Mueller issued a clear warning this week: bot-challenge screens ('Are You A Bot' interstitials) can cause pages to be dropped from Google's index, and may result in a different URL being selected as the canonical. This is a direct crawlability risk for sites using aggressive bot-protection layers — particularly relevant given the rise of AI scraper traffic prompting many site owners to implement new bot defenses, according to Search Engine Journal.

Google formalized long-standing verbal advice this week by updating its canonical help documentation to explicitly recommend including a self-referential canonical tag on the canonical page itself — a best practice previously communicated informally but now officially documented, per Search Engine Roundtable.

Separately, John Mueller addressed a nuanced A/B testing scenario. He confirmed that A/B tests with 'significantly different' content variants can surface in search results, and that depending on site setup, Google may index one variant over the other. In a seemingly contradictory statement covered by Search Engine Journal, Mueller also said year-long A/B tests won't result in SEO penalties — a claim that directly conflicts with Google's own official documentation, which warns such tests could be treated as deceptive cloaking. SEOs running extended CRO tests should weigh both signals carefully.

Mueller also addressed link obfuscation tactics tied to 'first-link priority' — where site owners hide homepage links hoping only a preferred anchor text counts. Mueller's response clarified Google's stance: the tactic's effectiveness is limited and the attempt to game internal link signals through hidden links is unlikely to produce meaningful results.

Google's Gary Illyes issued a direct and actionable technical SEO tip via Bluesky: if your XML sitemap lastmod dates are inaccurate, you're better off omitting them entirely. His exact words: *'probably better off without the lastmods. at least you save a few bytes.'* Incorrect lastmod timestamps can mislead Googlebot about when content was updated, potentially causing freshness signals to misfire. This is a concise but important callout for SEOs managing large sitemaps with automated date stamping, per Search Engine Roundtable.

Google made two significant moves in image search this week. First, AI image generation is now integrated directly into AI Overviews in Search — users can generate images from within the AI Overview response without leaving the results page. Google's announcement states: *'To help bring those unique ideas to life, we're bringing image generation directly into AI Overviews in Search,'* per Search Engine Roundtable and Search Engine Journal.

Simultaneously, Google overhauled the Google Image Search homepage for its 25th anniversary, replacing the classic minimalist search box with an image gallery layout and repositioning the search bar at the top. The UI change does not affect ranking algorithms but signals continued investment in image search as a surface.

Google's AI Mode continued its rapid expansion this week on two fronts. Connected Apps are now rolling out in AI Mode, allowing users to dispatch tasks directly to third-party services like Canva from within a search result — effectively turning search into a task execution layer rather than a navigation tool. The feature represents a meaningful reduction in the number of clicks that reach traditional websites, per Search Engine Journal.

In parallel, Google is expanding Personal Intelligence in AI Mode with Google Calendar integration, enabling the assistant to schedule events and factor a user's personal schedule into search responses. These updates together point to AI Mode's trajectory: less 'search and browse,' more 'instruct and execute.'

Google Merchant Center is rolling out a beta AI Performance report that gives retailers direct visibility into how their products perform within AI Mode and AI Overviews specifically. Google describes the report as *'specifically revealing your performance on AI Mode and AI Overviews'* — making it one of the first native reporting tools to quantify product visibility on Google's AI-powered search surfaces. This is a significant development for e-commerce SEOs who have had no direct way to measure AI surface performance until now, per Search Engine Roundtable.

Aleyda Solis published original data-backed research this week analyzing 15 leading SaaS brands across three verticals (CRM/Sales, and others) to identify which page types and third-party sources drive citations in ChatGPT and Google AI Mode. The study combines Semrush citation data with Similarweb AI referral data to map the full pipeline: which brand-owned pages get cited, which third-party sources reinforce those citations, and where measurable referral traffic actually lands.

The research provides an unusually complete picture of the AI citation ecosystem for SaaS, connecting citation patterns to actual traffic outcomes rather than treating citations as a vanity metric.

This week adds three new dimensions to the AI citation strategy conversation. First, Search Engine Journal's Roger Montti argues that content specificity is a key driver of AI citations — concrete data points, named examples, and precise claims outperform vague or generic content in AI-generated answers. Second, Eli Schwartz (The Future of SEO newsletter) warns that AI citations must reflect genuine real-world brand reputation rather than manufactured claims, arguing inflated or false citations will ultimately erode rather than build brand trust in AI environments — a counterpoint to the tactics-first GEO advice proliferating elsewhere. Third, a Search Engine Journal piece details Reddit's growing role as a sourcing layer for AI answers, particularly for multi-location businesses, outlining five specific plays for leveraging community-generated content. Moz's Whiteboard Friday with Charlie Marchant complemented this with a framework for proactive outreach to get brand mentions into the sources AI systems draw from. These pieces extend the proprietary data and authenticity themes from our June 29th coverage into more tactical territory.

Kevin Indig and David Kaufman of Siteline published original research analyzing how AI agents crawl and interact with B2B websites, identifying a critical failure point where agents get stuck and cannot complete tasks. The research is particularly timely given that Salesforce data shows 20% of sales now come from agents, 60% of companies have AI agents live in production, and the web now receives more bot traffic than human traffic.

The findings have direct implications for technical SEO readiness: sites optimized for human navigation may have structural elements — JavaScript-heavy interfaces, multi-step authentication flows, or non-semantic markup — that consistently trip up AI agents. SearchPilot and Sitebulb's complementary piece on 'Agentic Commerce SEO' reinforces the need for enterprise e-commerce teams to begin testing their sites against agentic traffic now.

Google rebranded NotebookLM to Gemini Notebook this week, renaming the associated crawler user agent from `Google-NotebookLM` to `Google-GeminiNotebook`. This is directly relevant to the broader bot management theme flagged in our June 22nd coverage of the PACT protocol and Cloudflare's AI crawler gatekeeping work. The practical urgency here is immediate: site owners who configured robots.txt rules targeting the old user agent name will find those rules no longer apply, potentially exposing their content to increased AI scraping without their intended restrictions. Publishers should update their robots.txt configurations to reference the new `Google-GeminiNotebook` user agent name immediately, per Search Engine Journal.

OpenAI updated its crawler documentation to clarify that the version number within its user-agent string may be updated, and that an additional robots.txt marker could be introduced. Site owners who use exact-match rules to block or permit OAI-SearchBot should audit their configurations for version-number dependencies, per Search Engine Roundtable. The change underscores a broader challenge: as AI crawlers proliferate and update their identifiers, static robots.txt rules require ongoing maintenance rather than a set-and-forget approach.

Following the confirmed conclusion of the June 2026 spam update — covered extensively in our June 22nd and June 29th issues — Search Engine Roundtable's Barry Schwartz is now flagging possible new ranking volatility he is calling the '7-Eleven update', observed around July 11th. Notably, community chatter about this potential update is at an all-time low compared to past unconfirmed updates, making the signals unusually quiet. No official Google confirmation has been issued. SEOs should monitor their tracking tools for pattern changes in the July 10–13 window.

Microsoft Bing's Principal Product Manager Krishna Madhavan made a notable transparency statement this week: Bing does not issue one-off manual penalties because it is not scalable. Instead, when Bing identifies a spam pattern, it builds a spam algorithm to detect that behavior at scale across the entire index — affecting all sites exhibiting that pattern simultaneously, not just the one that triggered the review. This contrasts directly with Google's manual actions system, where individual sites can receive targeted penalties. The statement has direct implications for how SEOs approach Bing compliance strategies, per Search Engine Roundtable.

A significant measurement gap was detailed by Search Engine Journal: GA4's built-in AI Assistant channel systematically undercounts AI referral traffic by splitting traffic from a single AI source into three separate channel buckets. This fragmentation makes AI-driven referral numbers 'quietly wrong' without users realizing it. The fix involves building a custom channel grouping in GA4 that consolidates AI traffic sources into a single, accurate view — the article provides a walkthrough for the implementation.

Google's John Mueller clarified what actually happens when the 'Validate Fix' button is clicked in Google Search Console — a commonly misunderstood feature. Mueller explained what process the button triggers under the hood and in which situations it is actually worth using versus when it may be premature. The guidance helps SEOs avoid triggering unnecessary recrawl requests before fixes are properly implemented, per Search Engine Journal.

Following last week's coverage of ChatGPT Ads' global expansion (Japan, South Korea) and the addition of AI-generated ad creative and audience targeting options, OpenAI has continued its rapid platform buildout this week with two further additions. Attributed Sales Value and Sales ROAS metrics now appear in the advertiser dashboard — the first true performance marketing accountability metrics on the platform, directly addressing the ROI measurement gap that has been the primary barrier to advertiser adoption. Additionally, geo-location and audience exclusion controls were added. Together, these updates represent a notable maturation step for ChatGPT as a paid channel competing with Google and Meta on measurable ROI, per Search Engine Roundtable.

Apple Maps updated its advertising policies to prohibit ads from home service businesses — including plumbers, electricians, HVAC providers, locksmiths, pest control, and roofing — as well as bail bonds, cryptocurrency ATMs, and certain medical services. The policy change is significant for local search advertisers in these verticals who had been preparing for Apple Maps Ads' broader rollout, per Search Engine Roundtable and Search Engine Journal. Affected businesses will need to redirect their local search advertising budgets to other platforms.

Two notable Google Shopping changes are incoming. First, Google will merge its Shopping Ads and Free Listings policies into a single consolidated document in September 2026 — an administrative consolidation but one that requires SEOs and merchants to re-familiarise themselves with combined compliance requirements. Second, and more immediately impactful, Google is enabling Local Inventory Ads by default for all Shopping campaigns, with an opt-out deadline of August 31st. Retailers who do not want local inventory included in their Shopping campaigns must switch from the 'Local products' setting to the 'inventory filter' setting before that date, per Search Engine Roundtable.

Industry veteran Avinash Kaushik made a provocative argument covered by Search Engine Journal: SEO agency fees could and should drop by 25–75% as AI tools shift the nature of SEO work from activity-based execution to judgment-based strategy. His central thesis is that contracts built around activity (hours, deliverables, audits) always reward more activity rather than better outcomes. His prescription: renegotiate agency scopes of work to be tied to verified revenue outcomes rather than task completion. The piece sparked debate about how agencies should position themselves as AI compresses the time required for traditional SEO tasks.

Publishers filed a class action lawsuit alleging Google used books from Google Books, Play Books, and Google Scholar to train its Gemini AI without permission. No court has yet ruled on the claims. The case could have significant downstream implications for AI training data practices and copyright law as they apply to content creators and publishers — a developing story worth monitoring for its potential to reshape how AI companies acquire training data, per Search Engine Journal.