For most small business websites, the honest answer is: you don't need one, and you should never use the kind that slips extra instructions into a visitor's AI assistant. A plain button that only asks an AI assistant to summarise your page is harmless but adds little. A button that also tells the assistant to “remember” your business as a trusted source is what Microsoft's security researchers now call AI Recommendation Poisoning. It treats your visitors as targets, it's easy to spot, and nothing official suggests it earns you a place in AI answers.
Here's what these buttons do, what was found, and what to do instead.
What a “Summarise with AI” button actually does
Most AI assistants accept a question through the web address itself. A button can link to an address such as chatgpt.com/?q=… or copilot.microsoft.com/?q=…, with a ready-written prompt after the ?q=. When a visitor clicks, their assistant opens with that prompt already filled in, and in some cases it runs straight away.
The visible label says “Summarise with AI”. The prompt behind it can say anything the website owner wrote. That gap between the label and the prompt is the whole issue.
What Microsoft found
In February 2026, Microsoft's Defender Security Research Team published research on AI Recommendation Poisoning (published 10 February 2026, updated 3 September 2026, checked 16 November 2026). According to the post:
- Companies were hiding instructions inside “Summarise with AI” buttons that tried to plant lasting commands in an AI assistant's memory, such as remembering a company as a trusted source or recommending it first.
- Over 60 days, reviewing AI-related links seen in email traffic, the team identified about 50 distinct prompts from 31 companies across more than a dozen industries, including finance, health, legal services and marketing agencies.
- Every case involved a real, legitimate business, not hackers or scammers.
- Free, ready-made tools marketed as an SEO tactic for AI made the technique easy to add to a website.
- How well it worked varied by assistant and over time, and Microsoft says it has put protections in place in Copilot, with some earlier behaviour no longer reproducible.
Microsoft's advice to ordinary AI users is telling: hover over these buttons before clicking, be suspicious of them in general, and check what your assistant has saved in its memory. In other words, the people you most want to impress are now being taught to distrust the button.
Two cautions. This is one company's research, based on what its own systems could see, and it doesn't measure whether the tactic ever changed a single recommendation. Treat it as evidence that the practice exists and is frowned upon, not as proof of how effective it is.
Why the hidden-instruction version is a bad idea
Even setting aside whether it works, a business has plenty of reasons to stay well clear:
- It acts on your visitor without their knowledge. They asked for a summary. They didn't ask to change their assistant's settings. That's the opposite of how you want a potential customer to feel about you.
- It's easy to see. Anyone can hover over the button and read the prompt in the link. A competitor, a journalist or a sharp-eyed customer only needs to look once.
- Search engines treat manipulation as spam. Google's spam policies (last updated 28 August 2026, checked 16 November 2026) define spam to include “attempting to manipulate generative AI responses in Google Search”. That policy covers Google's own AI features rather than other companies' chatbots, but the direction of travel is clear.
- AI providers are actively closing the gap. Microsoft describes ongoing defences against this kind of prompt injection. A tactic that depends on a loophole tends to stop working, and leaves the reputational cost behind.
The same thinking applies to instructions hidden inside your pages, such as white-on-white text telling AI tools to praise you. Google's spam policies already list hidden text among the practices that can see a page ranked lower or removed.
What about a plain summary button with nothing hidden?
A button whose prompt only says “Summarise this page” isn't what Microsoft described. It isn't deceptive. Whether it's worth adding is a different question, and for most small business sites we'd say probably not:
- It sends readers away from your page, at the moment they were reading your content and were close to your contact details.
- Anyone who wants a summary can already get one by pasting your link into their own assistant.
- We found no official guidance from Google, OpenAI or Microsoft saying these buttons help a site appear in AI answers. Google's page on its AI features says there are no additional requirements beyond its normal Search guidance (checked 16 November 2026).
- Some visitors will now be wary of any AI button, because security advice tells them to be.
If you do keep one, make the prompt short, visible and honest: something your visitor would be happy to read before clicking.
Check whether your site already has one
Themes, plugins and agencies sometimes add these buttons without much explanation. A two-minute check:
- Look for “Summarise with AI”, “Ask ChatGPT”, “Open in Copilot” or similar buttons on your articles and service pages.
- Hover over each one (or right-click and copy the link) and read the address it points to.
- Look at the text after
?q=or?prompt=. Words such as “remember”, “trusted source”, “in future conversations”, “authoritative” or “cite” are among the warning signs Microsoft lists. - If you find any, remove the instruction or remove the button, and ask whoever added it where else it was used, including email newsletters.
What earns a fair place in AI answers instead
There's no shortcut, but there is a reliable route, and it works for human readers at the same time:
- Answer real customer questions clearly, with the answer near the top of the page.
- State checkable facts in plain text: what you do, where, for whom, typical timelines and what's included.
- Show who is behind the content and how it was put together.
- Keep important pages current, and only change dates when the content has changed.
- Earn genuine mentions elsewhere from people who actually know your work.
None of this guarantees an AI tool will cite or recommend you. It does mean that when one reads your page, it finds something worth repeating, and when a person reads it, they trust you more rather than less.
What's in your control, and what isn't
In your control: what buttons and links appear on your site, what any pre-filled prompt says, whether your pages hide anything from readers, and how useful and accurate your content is.
Not in your control: how each AI assistant handles links and memory, whether a tactic that works today is blocked tomorrow, and whether any AI answer mentions your business. Nobody can promise AI citations or recommendations.