Search Report • July 5, 2026

Search Report by curious digital

The endings continued this week. Bing’s most visible search engineer retired after almost 30 years, and Google shut down the AMP cache it built a decade ago. The bigger stories are in the data: two new studies show AI is redrawing which sources get cited and how much demand reaches search at all.

Bing’s Fabrice Canel has retired

Fabrice Canel retired from Microsoft on July 1, after almost 30 years with the company. He led Bing’s crawling and indexing work, created the IndexNow initiative, and helped build Bing Webmaster Tools.

His departure removes the Bing side’s most visible contact for the SEO community. Microsoft has not announced who will take over webmaster communications. IndexNow and Bing Webmaster Tools continue as usual.

Sources: Search Engine Land, “Fabrice Canel retires from Microsoft Bing after legendary career”, Search Engine Journal, “Fabrice Canel, Longtime Bing Search Leader, Retires From Microsoft”

The AMP cache is gone

As of July 1, clicking an AMP result in Google Search takes you directly to the publisher-hosted AMP page. Google no longer serves a cached copy inside the AMP viewer with a google.com URL. Google also removed mentions of the AMP viewer, AMP Cache, and signed exchanges from its documentation.

Rankings are unchanged. Google told Search Engine Land that AMP content will continue to rank like any other webpage, and that publishers no longer need to update the AMP cache or configure signed exchanges. AMP

If your site still serves AMP pages, two checks are worth running this week. Confirm your AMP host pages load fast without Google’s CDN doing the work. Then ask the harder question: with Core Web Vitals doing what AMP was built for, is the format still earning its maintenance cost?

Sources: Search Engine Land, “Google Search now sends searchers directly to publisher-hosted AMP pages”, Search Engine Journal, “Google Ends Cache-Served AMP Pages In Search”

ChatGPT Thinking mode cites a different web

Semrush and Kevin Indig ran 100 prompts through GPT-5.2 twice, once with minimal reasoning and once with high reasoning. The two modes cited almost entirely different sources.

Only 25.6% of cited domains overlapped between modes for the same prompts.

Citation rates rose from 50% to 68% with high reasoning on. Sources per response nearly doubled, from 2.6 to 4.5. High reasoning ran 1,130 web searches across the test set. Minimal reasoning ran 245.

The source mix shifted too. Reddit’s citation share fell from 15% to 7% in Thinking mode. Government and academic sources rose from 1.9% to 8.8%, and official documentation gained ground.

The practical takeaway: ChatGPT is two different search surfaces wearing one interface. Visibility in fast default answers does not guarantee visibility when a user switches to Thinking mode for a complex question. If you track AI visibility, split your prompt tracking by reasoning mode. And if your citations lean on user-generated mentions, know that they lose roughly half their weight the moment the model thinks harder.

Sources: Semrush, “Only 25% of cited sources overlap between ChatGPT’s different reasoning modes”, Search Engine Land, “ChatGPT Thinking mode changes which brands get cited”

Search demand is falling faster than Gartner predicted

In 2024, Gartner predicted traditional search volume would fall 25% by 2026. Fractl and Search Engine Land tested that prediction against Semrush data for 1,010,848 keywords, each with 10,000 or more monthly searches, across 379 brands in eight verticals.

The result: 29% of high-volume search demand is in measurable decline. Four points worse than the forecast.

The damage is uneven. FinTech declined the most at -37.7%. Lifestyle declined the least at -15.2%. The pattern follows how information-heavy a category is. Where a chatbot can deliver a complete answer, such as an explanation of deductibles or a drug interaction summary, search volume drops.

For Canadian teams planning 2027 content budgets, this is the study to bring to the meeting. Audit which of your keywords are losing volume, identify the query patterns still growing, and shift effort toward the authority signals that get you cited in AI answers, not just ranked under them.

Source: Search Engine Land, “What 1 million keywords reveal about AI’s impact on search”

AI Mode recipe results now favour creators

On June 30, Google launched a new visual treatment in AI Mode that places prominent recipe links at the top of responses. The links include the creator’s name, recipe ratings, and the number of ingredients, with images. It builds on changes Google announced in March after sustained feedback from recipe creators.

Publishers are cautiously positive. Inspired Taste called the update a step in the right direction while repeating that the full AI-generated recipes below the cards still misrepresent, and sometimes plagiarize, their content.

This matters well beyond food blogs. It is the clearest signal yet that Google will redesign AI surfaces under publisher pressure, and the details surfaced in the new link units map directly to structured data fields like author and ratings. Clean, complete schema is what feeds these units. Keep yours in order.

Sources: Search Engine Land, “Google makes recipes in AI Mode more publisher friendly”, Search Engine Roundtable, “Google AI Mode Improves Links For Recipes, Again”