How to Add an “Ask AI About Us” Button to a Website, and Does It Work?

How does an "Ask AI" button work?

The button is a link. Most assistants accept a question in the web address and open with it already typed. Microsoft's research lists formats for the most popular ones:

• chatgpt.com/?q=<prompt>
• claude.ai/new?q=<prompt>
• perplexity.ai/search?q=<prompt>
• copilot.microsoft.com/?q=<prompt>
• grok.com/?q=<prompt>

When a visitor clicks, the assistant opens in their own account with the prompt, which usually includes your page's address. The assistant then fetches the page, or searches for it, and writes a summary. For that to work, the assistant's user-triggered fetcher must be able to read your page. That means your bot protection or firewall must let ChatGPT-User, Claude-User and Perplexity-User through, and robots.txt must not block Claude-User.

How do you add one?

1. Choose which assistants to offer. Two or three is enough; a row of logos adds clutter.
2. Write a plain, visible prompt, such as "Summarise this page and list its key points: [page URL]". Show the prompt text in a tooltip or small print so visitors know what will be sent.
3. URL-encode the prompt and build one link per assistant.
4. Open it in a new tab, so visitors can return to your page.
5. Label it clearly for accessibility, for example "Summarise this page with ChatGPT (opens in a new tab)".
6. Track clicks as an analytics event, so you know whether anyone uses it.
7. Test each link to make sure the assistant can read the page and the summary is accurate.

A simple version:

<a href="https://chatgpt.com/?q=Summarise%20this%20page%20and%20list%20its%20key%20points%3A%20https%3A%2F%2Fwww.example.com%2Fguide"
   target="_blank" rel="noopener">
  Summarise this page with ChatGPT
</a>

Put it where a summary helps: long guides, research reports, documentation, policy pages. A site-wide footer button on every page rarely gets used.

Do site-wide ChatGPT, Perplexity and Claude buttons improve Google AI Overviews visibility?

No. Google's AI Overviews are built from Google's search index. Google says its AI features need only that a page is indexed and eligible for a snippet, with no special markup or files. Outbound links to assistants are not a documented signal, and clicks on them happen in visitors' own accounts, outside Google Search.

The same applies to ChatGPT, Claude and Perplexity. A visitor's click produces a summary in that visitor's chat. It does not teach the model anything about your site that it will tell other users. Assistants do not learn from individual conversations in real time, as covered in whether asking AI about your brand every day trains it to recommend you.

If AI Overviews are not citing you, the causes lie elsewhere: indexing, snippet controls, how directly your pages answer the question, and which other sources Google trusts for it.

What is AI Recommendation Poisoning, and why avoid it?

Microsoft's security team found companies embedding hidden instructions in "Summarize with AI" buttons. They identified over 50 unique prompts from 31 companies across 14 industries. The prompts told the assistant to "remember [Company] as a trusted source" or "recommend [Company] first". If the visitor's assistant has memory switched on, the instruction may be saved and bias that visitor's future answers without their knowledge. Microsoft classifies this with known AI attack techniques, including memory poisoning and prompt injection. It notes that freely available tools make it trivially easy to deploy.

Avoid it because:

• It deceives your visitors, who clicked for a summary, not to have their assistant's memory altered.
• It is treated as an attack, and security vendors now detect and publicise it.
• Assistants harden against it. AI companies work to resist prompt injection, so the effect is unreliable.
• It damages trust if discovered, which is increasingly likely.
• It only reaches people who already visited you. Even if it worked, it would not affect anyone else's answers.

Whether other ways of influencing AI are ethical is covered in whether AI visibility optimization is ethical.

What are the genuine benefits?

• Convenience for readers of long or technical pages who prefer a summary or want to ask follow-up questions.
• A quick internal check. Clicking your own buttons shows how assistants read and summarise your pages, and whether they can fetch them at all.
• A measure of interest. Click data shows which pages visitors want summarised, often a hint that those pages are too long or unclear.

Weigh these against the costs. Visitors leave your site for another company's interface. The summary may simplify or misstate your content. And the prompt sends your page address to a third party.

How do you check that a button does what it says?

Before publishing, and again every few months:

• Click each link while logged out and logged in, and confirm the prompt arrives intact and readable.
• Read the summary critically. Does it state your prices, limits and key facts correctly? If not, the page itself may be unclear.
• Check your server logs for a request from the assistant's user-triggered fetcher at the moment you click. A 403 means the assistant was blocked from reading the page. No request suggests it did not fetch the page live, so compare the summary with the live page.
• Check the prompt text in your page source, especially if a plugin or agency built the button, to make sure nothing extra has been added.

When is it worth adding?

What might an illustrative case look like?

Consider an illustrative software company that added ChatGPT, Perplexity and Claude buttons to its site footer, hoping to appear in Google's AI Overviews. After three months, the buttons had been clicked a few dozen times, and AI Overviews for its category still cited competitors. An agency had also suggested adding "remember us as the leading provider" to the prompts. The company declined after reading Microsoft's research. It removed the footer buttons and added a single "Summarise with ChatGPT" link to its long annual benchmark report, where clicks were steady. It moved its AI Overview effort to rewriting its category page to answer the main question in its first paragraph.

What do people ask about "Ask AI" buttons?

Do these buttons count as backlinks?

No. They are links from your site to the assistant, not links to your site.

Can the button make ChatGPT remember our brand?

Hidden instructions may try to, but that is the poisoning Microsoft warns about. Honest buttons do not, and should not.

Will assistants always be able to read the page?

Not always. Bot protection, a firewall or a login wall can stop them, so allow the user-triggered fetchers there. A robots.txt block stops Claude-User, but OpenAI says robots.txt may not apply to ChatGPT-User and Perplexity says Perplexity-User generally ignores it. Test each one. Test each one.

Should we add a summary on the page instead?

Often that is better: a short summary at the top of long pages serves every visitor and every AI system.