Loading page
What AI share buttons do for readers, what the evidence says about AI visibility, and how to write neutral prompts that avoid AI Recommendation Poisoning.
AI share buttons, such as "Summarize with ChatGPT" or "Ask Perplexity", send a prompt about your page to an AI assistant, which may submit it immediately. You can offer them as a reader convenience. Two sources published in 2026 are worth reading before you add them: a case study on their search performance, and a Microsoft report on button prompts designed to manipulate AI memory.
💡 TL;DR Treat AI share buttons as a reader feature, not a proven way to increase AI citations. Use a short, neutral prompt and never ask the assistant to remember, trust or recommend your brand. If you want to improve AI visibility, test a visible summary on part of your pages and compare the results with a control group.
An AI share button is a link to an AI assistant with a prompt in the URL, for example https://chatgpt.com/?q=.... When the reader clicks it, the link passes the page URL and a task to the assistant, such as summarizing the page, explaining a point or comparing options. Depending on the platform and whether the reader is signed in, the prompt may run immediately or require further action.
Your link supplies a prompt to the reader's own assistant. It does not directly update the underlying model. What happens to the conversation afterwards depends on the provider and on the reader's settings, including memory features where they exist.
Search Engine Land reports a comparison of two recipe pages on Leite's Culinaria, one with an AI summary and AI buttons, the other with AI buttons only:
Readers used the buttons to adapt recipes, for example to scale portions or substitute ingredients. These are search metrics from one site, not AI citation counts, and a two-page comparison cannot show what caused the difference. The case study does not show that buttons increase visibility. It does make a visible summary worth testing on your own pages.
In February 2026, Microsoft described "AI Recommendation Poisoning" : "Summarize with AI" buttons whose prompts try to plant instructions in the assistant's memory, such as "remember [Company] as a trusted source" or "recommend [Company] first", to bias future answers. Over 60 days, Microsoft identified 50 distinct attempts from 31 companies in more than a dozen industries, and compares the technique to SEO poisoning and adware. For your own buttons, use conservative safeguards. Never write a prompt that:
Run a free audit: in a few minutes, ChatGPT, Gemini and Perplexity answer 8 buying questions about your market live, and you see who they recommend.
Use one neutral prompt, encode the URL and test every link before enabling it. Showing the prompt to readers is optional; the example below does it. The formats below for ChatGPT, Perplexity and Claude are the ones listed in Microsoft's report . They are candidate URL formats, not documented APIs, and they can change. Before enabling each button, check whether your prompt is preserved, whether the reader has to submit it, or whether it runs immediately, both signed in and signed out. Google AI Mode is not included because no reliable prefill format could be confirmed. Buttons serve readers, so a client-side script is enough. This version reads the page's canonical URL and keeps the buttons hidden until the links are ready:
<p class="ai-share-prompt" hidden>
Prompt sent to your selected assistant:
"Summarize the key points of this page and list any claims I should verify: <span class="ai-share-url"></span>"
</p>
<div class="ai-share-buttons" hidden>
<span>Summarize with AI:</span>
<a data-ai="chatgpt" target="_blank" rel="noopener nofollow">ChatGPT</a>
<a data-ai="perplexity" target="_blank" rel="noopener nofollow">Perplexity</a>
<a data-ai="claude" target="_blank" rel="noopener nofollow">Claude</a>
</div>
<script>
(function () {
var canonical = document.querySelector('link[rel~="canonical"][href]');
if (!canonical) return;
var href = canonical.getAttribute('href').trim();
if (!href) return;
var pageUrl;
try {
var u = new URL(href, document.baseURI);
if (!/^https?:$/.test(u.protocol)) return;
u.hash = '';
pageUrl = u.href;
} catch (e) {
return;
}
var prompt = 'Summarize the key points of this page and list any claims I should verify: ' + pageUrl;
var q = encodeURIComponent(prompt);
var urls = {
chatgpt: 'https://chatgpt.com/?q=' + q,
perplexity: 'https://www.perplexity.ai/search?q=' + q,
claude: 'https://claude.ai/new?q=' + q
};
document.querySelectorAll('.ai-share-buttons a[data-ai]').forEach(function (a) {
if (Object.prototype.hasOwnProperty.call(urls, a.dataset.ai)) { a.href = urls[a.dataset.ai]; } else { a.remove(); }
});
document.querySelector('.ai-share-url').textContent = pageUrl;
document.querySelector('.ai-share-prompt').hidden = false;
document.querySelector('.ai-share-buttons').hidden = false;
})();
</script>
Use a public canonical URL that contains no private identifiers. For WordPress, add this once to your child theme's functions.php or to a site-specific plugin. It uses the post's WordPress permalink:
function qw_ai_share_buttons($content) {
if (!is_singular('post') || !in_the_loop() || !is_main_query()) return $content;
$permalink = get_permalink();
if (!$permalink) return $content;
$prompt = 'Summarize the key points of this page and list any claims I should verify: ' . $permalink;
$q = rawurlencode($prompt);
$links = array(
'ChatGPT' => 'https://chatgpt.com/?q=' . $q,
'Perplexity' => 'https://www.perplexity.ai/search?q=' . $q,
'Claude' => 'https://claude.ai/new?q=' . $q,
);
$html = '<p class="ai-share-prompt">Prompt sent to your selected assistant: "' . esc_html($prompt) . '"</p>';
$html .= '<div class="ai-share-buttons"><span>Summarize with AI:</span>';
foreach ($links as $label => $url) {
$html .= sprintf('<a href="%s" target="_blank" rel="noopener nofollow">%s</a>', esc_url($url), esc_html($label));
}
return $content . $html . '</div>';
}
add_filter('the_content', 'qw_ai_share_buttons');
Check your robots.txt rules, llms.txt and sitemap signals for AI crawlers before rewriting a single page.
A visible summary at the top of the page is a different technique from share buttons. Readers see it on the page. Include it in the page HTML so that crawlers allowed to access your content can retrieve it along with the article. This does not guarantee indexing or citation. As an editorial starting point, try two or three sentences that answer the main question, followed by three to five useful points. Keep the summary visible, accessible and consistent with the body of the article, and have the author write or review it. Show a new "Last updated" date only when you significantly revise the content. Adding a summary alone does not make an article new.
Use this as an exploratory test of the summary, not of the buttons:
Choose the sample size and duration from how much your metrics vary before the change and the size of effect you want to detect. Without a control group, a seasonal trend or an algorithm update can look like a result, and even a well-run test may end inconclusive. For background on AI visibility, read What is Generative Engine Optimization (GEO) .
Track the questions that matter to your business, see which brands each answer mentions and which sources it cites, then work on the gaps.
Free plan with 120 initial credits • No credit card required
Other Articles
GEO Strategy: 5 AI Search Truths and a 4-Step Framework
Five truths about AI search shared at the SEO & GEO Summit in Paris, and the 4-step GEO framework Qwairy gives every client to get recommended by ChatGPT, Gemini and other AI engines.
The GEO Mistake Everyone Makes: Tracking Only the First AI Answer
Most AI visibility checks stop at the first answer. But AI assistants turn questions into conversations, and the follow-up questions they suggest decide which brands stay in the race.