Microsoft is testing a change to Copilot Search that makes only the small citation marker at the end of an AI-generated sentence clickable, rather than the entire line. The shift, spotted by search observer Sachin Patel and reported by Search Engine Roundtable on April 10, 2025, may alter how millions of users verify information and could punch a hole in publisher traffic.

A tiny target with big consequences

Until now, clicking anywhere on a cited sentence inside Copilot Search would open the source. That design encouraged casual fact-checking: a quick tap on a claim about a Windows update or a restaurant review sent you to the original page. Under the reported test, the clickable zone shrinks to a superscript-style citation marker — a few pixels that are easy to miss on touch screens, high-resolution laptops, and for anyone with motor or vision impairments.

The change may look like a minor interface polish, but it attacks the last physical bridge between an AI answer and the open web. When the whole sentence was a link, users could verify a statement without hunting. Now verification demands precision. On a phone, aiming for a tiny number while scrolling is a chore; on a desktop, the cursor must land exactly on the marker. The behavioral effect is plain: harder-to-click links get clicked less.

Search Engine Roundtable could not replicate the test, suggesting it is a limited experiment. That offers small comfort. If the design spreads, the implications cascade.

What the change really does

Microsoft has not publicly described the test, and the company did not respond to a request for comment before publication. However, Patel’s screen recording, published on X, shows the old and new behaviors side by side. In the standard interface, hovering over any part of a sourced statement highlights the whole line in a link color. In the new version, only the bracketed number at the end acts as a hyperlink.

The switch alters two fundamental things:

  1. Affordance: A full-line link screams “click me.” A tiny marker does not. Many users may not even realize the number is interactive.
  2. Hit area: Fitts’s Law — the time to acquire a target is a function of its size and distance — predicts that smaller targets increase error rates and decrease selection speed. The citation marker can be 10 to 20 pixels wide; a full sentence might be 500 or more.

On touch devices, the problem deepens. Microsoft’s own Surface Pro or Android phones running Edge would struggle with a 12-pixel tap target. Accessibility tools like screen magnifiers or switch controls amplify the difficulty.

What it means for you

For everyday users

If you use Bing to check news, recipes, health advice, or tech support, the new design makes source-checking an afterthought. An AI summary might confidently declare that a Windows 11 update fixes a specific bug, but without an easy click to Microsoft’s support article, you won’t know if the fix applies to your build or if a manual workaround is still necessary. The barrier to verification rises, and with it, the risk of acting on incomplete or stale information.

Mobile is particularly affected. Thumb-zone analysis shows that small targets near the edge of a description are physically awkward to hit. A line-long link, by contrast, is forgiving. The practical result: more zero-click sessions where users accept the AI answer and move on, even when a deeper check would be healthier.

For publishers and content creators

Every click from a search engine sustains ad revenue, affiliate income, subscriptions, and brand visibility. When Copilot Search ingests your article and summarizes it, the citation remains your sole commercial lifeline. Shrinking the clickable area to a footnote-style marker likely depresses outbound traffic. A site may still appear as a source — the brand gets a mention — but visits decline.

This matters most for niche sites that depend on search referrals. A small technology blog, a local newsroom, or a medical reference page cannot survive on “exposure” alone. If Bing’s test expands, publishers should prepare for a world where AI visibility does not equal traffic, and where being cited feels more like being used.

Search Engine Roundtable’s own report illustrates the irony: the article you’re reading now might be cited by Copilot Search, yet the new design could prevent you from clicking through to the original.

For IT professionals, developers, and researchers

WindowsForum readers often search for error codes, PowerShell syntax, or driver conflicts. In these cases, the source is not a nice-to-have; it’s the only authority. A Copilot summary might correctly describe a known issue but omit crucial context like a prerequisite update or a registry caveat. The user must reach the Microsoft Learn document, the KB article, or the community forum thread instantly. A smaller click target slows that journey and increases troubleshooting errors.

The same applies in enterprise environments where auditability matters. A financial analyst verifying a regulatory claim or a security engineer checking a CVE detail cannot afford to guess where the AI got its information. If Copilot Search trains workers to accept answers without clicking through, corporate governance takes a hit.

How we got here

Microsoft launched the new Bing with AI in February 2023, branding it as a “copilot for the web.” From day one, source citations were central to the pitch. Satya Nadella talked about “grounding” answers in fresh web data, and the interface showed footnotes that expanded into source cards. The message was clear: Bing’s AI helps you explore, not just consume.

Over time, the product has been renamed, merged with the Copilot brand, and integrated across Windows, Edge, and Microsoft 365. But the citation mechanism remained largely unchanged — until now. The current test arrives at a moment when the entire search industry is wrestling with how to balance AI convenience against publisher survival.

Google’s AI Overviews have faced fierce criticism for reducing click-through rates, and the company has tweaked link placements multiple times. Perplexity has built a loyal following by making sources prominent and easy to open. Microsoft, with a smaller but commercially important search share, has a chance to differentiate itself by being the most publisher-friendly AI search engine. This test, however, pulls in the opposite direction.

What you can do now

If you’re a regular user

  • Be deliberate about clicking citations. When Copilot gives you an answer, treat the source markers as essential, not optional. If they’re too small, try zooming or use a mouse where possible.
  • Consider alternative search tools. If fact-checking matters to you, explore search engines that prioritise visible source links, such as DuckDuckGo, or AI tools like Perplexity that make citations a first-class interaction.
  • Send feedback to Microsoft. Use the feedback button inside Bing (the bottom-right smiley face) to report that smaller citation targets harm your experience. Product teams do read these inputs.

If you’re a publisher, SEO, or webmaster

  • Audit your Bing traffic. Look at your analytics for referrals from “bing.com/copilot” or “bing.com/chat”. Monitor changes in click-through rates if the test expands.
  • Double down on being the best source. AI models favour pages that are clearly authoritative, well-structured, and original. Ensure your content uses concise headings, answers questions directly, and includes unique data that an AI summary can’t fully replicate.
  • Diversify your traffic sources. Strengthen direct visits via newsletters, apps, and social media. Relying solely on any single search engine is risky when AI summaries are reshaping discovery.
  • Join the conversation. Publisher advocacy groups and SEO forums are tracking these changes. Your voice adds weight to calls for transparent, traffic-respecting AI design.

If you’re a Microsoft insider or an IT decision maker

  • Raise the issue internally. If you work with Microsoft products, emphasise to your account teams that enterprise users need reliable, easy-to-access sources in Copilot experiences. Both consumer and commercial Copilot share interface patterns.
  • Test the behavior if you can. If you have access to the experimental layout, document how it affects your workflows. Concrete examples help product groups understand real-world impact.

What happens next

Microsoft often runs interface experiments that never ship. The current test could be scrapped or modified based on telemetry and feedback. However, the very existence of the test signals a design philosophy that prioritises clean interfaces over clickable source pathways. Rivals will interpret it as a retreat from verification-friendly AI search.

Watch for these developments:

  • Whether Microsoft issues a statement explaining the rationale for the experiment.
  • Whether the smaller citation target appears on mobile devices, where the impact would be most severe.
  • Any response from major publishers, advertising trade groups, or regulators concerned about platform power over content.
  • How Google and other AI search engines adjust their own link designs in parallel.

The deeper question isn’t about a single UI element. It’s about whether AI search engines will value the web as a living ecosystem or treat it as a training corpus. A tiny click target says more than a press release ever could.