Which AI model invents the most URLs? 26,435 citations examined

<p><em>We regularly publish aggregated, anonymised figures from our platform. Here, a simple question no one answers with real data: when an AI cites a source, does that link actually exist? And which model gets it wrong most often?</em></p> <h2>The scope</h2> <p>Between September 2025 and July 2026, the platform analysed <strong>7,774 AI answers</strong> from <strong>132 audits</strong>, produced by <strong>53 distinct models</strong> (ChatGPT, Claude, Gemini, Perplexity, Grok, Mistral and others). Within these answers, the models cited <strong>26,435 URLs as sources</strong>. We checked every one: <strong>70 did not exist</strong> — plausible links to pages that were never published, spread across <strong>23 different brands</strong>. An average rate of <strong>0.26%</strong>, roughly <strong>1 cited URL in 380</strong>.</p> <p>The method is deliberately hard to dispute: a URL is counted as "invented" only if the model cited it as a source <em>and</em> it returns an error (404 or soft-404) at verification time. No interpretation: a page exists, or it does not.</p> <h2>The ranking — invented-URL rate per model</h2> <p>The only honest figure is not the raw error count, but the <strong>rate</strong>: what share of the URLs a model cites fails to resolve. Because a model that cites many links (like Perplexity) mechanically has more chances to be wrong. So here are invented URLs relative to total URLs cited, for each model with enough data.</p> <div style='overflow-x:auto;margin:20px 0;'><table style='width:100%;border-collapse:collapse;font-size:0.92rem;min-width:520px;'><thead><tr><th style='text-align:left;padding:10px 12px;border-bottom:2px solid #e5e7eb;font-size:0.85rem;'>Model</th><th style='text-align:left;padding:10px 12px;border-bottom:2px solid #e5e7eb;font-size:0.85rem;'>Invented-URL rate</th><th style='text-align:left;padding:10px 12px;border-bottom:2px solid #e5e7eb;font-size:0.85rem;'>Invented URLs</th><th style='text-align:left;padding:10px 12px;border-bottom:2px solid #e5e7eb;font-size:0.85rem;'>Brands affected</th><th style='text-align:left;padding:10px 12px;border-bottom:2px solid #e5e7eb;font-size:0.85rem;'>URLs cited</th></tr></thead><tbody><tr style='background:#faf5ff;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;'>Perplexity Sonar Pro</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>1,05 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>21</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>9</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>2 002</td></tr><tr style='background:#f9fafb;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:500;'>Anthropic Claude Sonnet 4.6</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>0,62 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>9</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>5</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>1 449</td></tr><tr style='background:#ffffff;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:500;'>OpenAI GPT-5 Mini</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>0,44 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>24</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>13</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>5 436</td></tr><tr style='background:#f9fafb;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:500;'>Anthropic Claude Haiku 4.5</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>0,19 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>8</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>3</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>4 198</td></tr><tr style='background:#ffffff;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:500;'>OpenAI GPT-5</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>0,16 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>3</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>2</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>1 848</td></tr><tr style='background:#f9fafb;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:500;'>Google Gemini 2.5 Flash</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>0,12 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>2</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>1</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>1 656</td></tr><tr style='background:#ffffff;'><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:500;'>Perplexity Sonar</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-weight:700;font-variant-numeric:tabular-nums;'>0,04 %</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>3</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>3</td><td style='padding:10px 12px;border-bottom:1px solid #f0f0f0;font-variant-numeric:tabular-nums;color:#6b7280;'>7 065</td></tr></tbody></table></div> <p style="font-size:0.9em;color:#6b7280;">How to read it: of the 2,002 URLs cited by Perplexity Sonar Pro in our audits, 21 (1.05%) did not resolve, across 9 brands. Models with 3 invented URLs or fewer are indicative — the sample is too small to rank finely, and we say so.</p> <h2>Finding 1 — The "search engine" paradox</h2> <p>The model that invents the most URLs, proportionally, is <strong>Perplexity Sonar Pro</strong> — precisely the one sold as a search engine that cites its sources. Yet the same vendor's lighter version (<strong>Perplexity Sonar</strong>) posts the <strong>best</strong> score of the whole panel (0.04%). The gap between the two is about <strong>25×</strong>. The lesson: citation reliability doesn't depend on the model's brand, but on its precise configuration. That is exactly why measuring <a href="/en/ags-methodology">model by model</a> matters — a generic "ChatGPT" or "Perplexity" means nothing.</p> <h2>Finding 2 — In absolute terms, the most-used model produces the most errors</h2> <p><strong>GPT-5 Mini</strong> produces the highest raw count of invented URLs (24, across 13 brands) — not because it is the least reliable (its 0.44% rate sits mid-table), but because it is by far the most-queried model in our audits. In other words: at scale, even an average rate yields a substantial volume of errors. That is the trap of "default" models.</p> <h2>Finding 3 — No model is immune</h2> <p>Every family in our data — OpenAI, Anthropic, Google, Perplexity — produced at least one invented URL. There is no "clean" model at 0%. And a URL invented by a citation-first AI is <strong>more dangerous</strong> than a plain error: the user trusts the displayed source. In our <a href="/blog/en/ai-visibility-barometer-brands-2026">2026 barometer</a>, we found 8 of these phantom addresses that received real human visits — prospects clicking and landing on an error page attributed to your brand. It is a reputation blind spot, detailed in our <a href="/blog/en/hallucinated-urls-when-ai-invents-pages-on-your-website">hallucinated URLs</a> guide and the <a href="/glossaire/en/ai-hallucination">AI hallucination</a> term.</p> <h2>What it changes for a brand</h2> <p>Two concrete takeaways. First, the model you are visible in matters as much as being visible: a citation in a model that invents links exposes you to erroneous associations. Second, the only way to know is to measure per model and verify every cited URL — which no manual monitoring can do at this scale.</p> <h2>The limits of these figures</h2> <p>This data comes from brands that chose to audit themselves — a sample that is not representative of all companies. Models with a low error volume (3 URLs or fewer) are indicative, not definitive. Rates evolve with every model update; we will republish this study with fresh data. The full methodology, including what we do not measure, is on our <a href="/en/ags-methodology">methodology page</a>.</p> <p>Want to know which models cite — or invent — URLs for <em>your</em> brand? That is exactly what an audit measures, model by model, URL by URL.</p>