---
title: "AI Engines Name Only 18% of the Sources They Cite: ChatGPT 50%, Gemini 17%, Claude 13%, Perplexity 8%"
description: "In 7,098 source citations across 899 AI answers, 17.6% were named in the answer text: ChatGPT 49.9%, Gemini 17.1%, Claude 13.1%, Perplexity 8.0%."
canonical: "https://machinerelations.ai/research/ai-engines-cite-sources-they-do-not-name-2026"
---

# AI Engines Name Only 18% of the Sources They Cite: ChatGPT 50%, Gemini 17%, Claude 13%, Perplexity 8%

In 7,098 source citations across 899 AI answers, 17.6% were named in the answer text: ChatGPT 49.9%, Gemini 17.1%, Claude 13.1%, Perplexity 8.0%.

Canonical URL: https://machinerelations.ai/research/ai-engines-cite-sources-they-do-not-name-2026
Published: 2026-11-07
Research type: Index Analysis
Tags: citations, brand-mentions, attribution, measurement, machine-relations-index

## Current Index data

Updated from the Index release through **2026-11-06** (mri_score_v2.1); observation window 2026-05-10–2026-11-06. This card updates with the published dataset. The article below remains a dated analysis; these figures do not retroactively validate its claims.

- [forbes.com](https://machinerelations.ai/index/domains/forbes.com): 3.40% citation rate; 1,249 of 36,708 observed runs, tier A.
- [techradar.com](https://machinerelations.ai/index/domains/techradar.com): 2.79% citation rate; 1,023 of 36,708 observed runs, tier A.
- [arxiv.org](https://machinerelations.ai/index/domains/arxiv.org): 2.49% citation rate; 913 of 36,708 observed runs, tier A.
- Engines named here and observed in this dataset: ChatGPT, Claude, Gemini, Perplexity. Coverage varies by engine; see the [release manifest](https://machinerelations.ai/data/mri-release-manifest.json) for returned runs and availability.

Citation presence measures observed source selection, not endorsement or causation. [Explore the Index](https://machinerelations.ai/index) · [Read the latest weekly report](https://machinerelations.ai/research/machine-relations-index-report).

## Source Body

AI engines cite far more sources than they name. In the Machine Relations Index answer ledger, 899 answers from four engines to 49 AI-visibility buying questions cited 7,098 source-and-answer pairs that could be checked, and the answer text named the cited source in 1,250 of them, or 17.6 percent (95% interval 13.1 to 21.7). The rate depends on the engine: ChatGPT named 49.9 percent of the sources it cited, Gemini 17.1, Claude 13.1 and Perplexity 8.0.

A citation and a mention are two different events. A source can sit in the citation list of an answer that never says its name, and a name can appear in an answer that cites a different page. This page counts how often the two coincide.

## What was counted

- **Panel.** The 49 questions of the Index's AI-visibility category (best tools, how to choose, is it worth it, problem-first research, top lists and head-to-head comparisons). The same answers as in [What AI Engines Call This Work](/research/what-do-ai-engines-call-ai-visibility-work-term-share-2026): 1,129 successful runs on 23 dates from August 7 to November 2, 2026, with ChatGPT 295, Claude 295, Perplexity 295 and Gemini 244.
- **A cited source.** A root domain in the run's stored citation events, counted once per answer however many of its pages were cited. 908 of the 1,129 answers cited at least one source, in 7,807 distinct source-and-answer pairs.
- **Named.** The answer text, with links and domain-style tokens removed, contains the source's brand name: the first label of its domain, with the prefixes "try" and "get" dropped where six or more letters remain, matched as a whole word and allowing a space between letters (so "SE Ranking" matches seranking.com).
- **Pairs left out.** 372 pairs whose name is under five letters, 274 pairs from nineteen general platforms (Google, YouTube, Reddit, LinkedIn, Wikipedia and similar, whose names appear in answers for reasons that have nothing to do with the cited page) and 63 pairs where the question itself contained the source's name, so the answer could not help saying it. That leaves 7,098 pairs from 1,758 domains in 899 answers.
- **Uncertainty.** The 95 percent intervals resample the 49 questions, each question's answers kept together, 2,000 times. Answers to one question are not independent, so intervals run wider than a count of pairs would suggest.

## How often the cited source is named

| Engine | Answers with a testable cited source | Cited sources per answer | Named per answer | Cited sources named | Share | 95% interval |
|---|---:|---:|---:|---:|---:|---:|
| ChatGPT | 255 | 4.7 | 2.3 | 598 of 1,199 | 49.9% | 38.7 to 59.5 |
| Gemini | 115 | 11.2 | 1.9 | 221 of 1,290 | 17.1% | 12.8 to 21.5 |
| Claude | 234 | 5.2 | 0.7 | 158 of 1,206 | 13.1% | 8.6 to 17.9 |
| Perplexity | 295 | 11.5 | 0.9 | 273 of 3,403 | 8.0% | 4.9 to 11.0 |
| All four | 899 | 7.9 | 1.4 | 1,250 of 7,098 | 17.6% | 13.1 to 21.7 |

Perplexity cited the most sources and named the fewest: 11.5 per answer, 0.9 of them by name. ChatGPT cited fewer than half as many and named more than two of them. In 466 of the 899 answers, 51.8 percent, none of the testable cited sources was named at all. Of 1,758 distinct cited domains, 422 were named in at least one answer.

Only 115 of Gemini's 244 answers carried a citation list in the stored events, so its row rests on fewer answers than the others.

## Which sources get named

The table lists every domain cited in at least 25 answers, from most often named to least. Most of the top rows are products that a buying question could be recommending.

| Source | Answers citing it | Answers naming it | Share |
|---|---:|---:|---:|
| tryprofound.com | 100 | 69 | 69.0% |
| brand24.com | 29 | 14 | 48.3% |
| scrunch.com | 30 | 14 | 46.7% |
| ahrefs.com | 56 | 24 | 42.9% |
| semrush.com | 129 | 54 | 41.9% |
| writesonic.com | 45 | 18 | 40.0% |
| otterly.ai | 60 | 22 | 36.7% |
| siftly.ai | 37 | 12 | 32.4% |
| seranking.com | 89 | 28 | 31.5% |
| similarweb.com | 33 | 9 | 27.3% |
| deepsmith.ai | 28 | 7 | 25.0% |
| llmpulse.ai | 63 | 15 | 23.8% |
| hubspot.com | 177 | 34 | 19.2% |
| frase.io | 42 | 8 | 19.0% |
| presenc.ai | 27 | 5 | 18.5% |
| yotpo.com | 48 | 8 | 16.7% |
| aiclicks.io | 25 | 4 | 16.0% |
| onely.com | 25 | 4 | 16.0% |
| authoritytech.io | 32 | 5 | 15.6% |
| airops.com | 54 | 8 | 14.8% |
| adobe.com | 25 | 3 | 12.0% |
| dageno.ai | 47 | 4 | 8.5% |
| zapier.com | 35 | 2 | 5.7% |
| techradar.com | 60 | 2 | 3.3% |
| searchengineland.com | 39 | 1 | 2.6% |
| nicklafferty.com | 58 | 1 | 1.7% |
| rankability.com | 25 | 0 | 0.0% |
| averi.ai | 38 | 0 | 0.0% |
| higoodie.com | 28 | 0 | 0.0% |
| discoveredlabs.com | 34 | 0 | 0.0% |
| stackmatix.com | 68 | 0 | 0.0% |
| therankmasters.com | 61 | 0 | 0.0% |
| beomniscient.com | 38 | 0 | 0.0% |
| slatehq.com | 34 | 0 | 0.0% |
| arxiv.org | 34 | 0 | 0.0% |

Two further domains with 25 or more citations, trysight.ai and gracker.ai, are left out of the table because the rule found none or almost none of their names and at least one is written another way: a Claude answer wrote "grackerai" for gracker.ai, which the whole-word match does not count. They are in the totals, so every figure on this page is a floor, not an exact count.

The pattern in this panel is that the sources the questions are about get named and the sources that explain things mostly do not. Profound, Semrush, Ahrefs and the other trackers and platforms are what a question about AI-visibility tools is looking for. TechRadar was cited in 60 answers and named in 2, Search Engine Land in 39 and named in 1, and arXiv in 34 and named in none. This is what the panel shows. It is not tested: the questions were written about tool buying, and a panel about something else could look different.

## Where AuthorityTech and Machine Relations Research stand

Machine Relations Research, which publishes this page and the Index, and AuthorityTech are both owned by Bramwell, Inc. Their rows are counted by the same rule as everyone else's.

- **authoritytech.io** was cited in 32 answers on all four engines and named in 5: Claude 2 of 12, Gemini 2 of 5, Perplexity 1 of 11, ChatGPT 0 of 4. At each engine's pooled rate it would have been named in about 5.3. It sits where the engine mix says it should.
- **machinerelations.ai** was cited in 3 testable answers and named in none, and **jaxonparrott.com** in 1 and named in none. Those numbers are too small to say anything beyond their size.

Being cited in these answers did not carry a source's name with it, ours included. It did not carry the term either: of the 36,836 ledger answers to questions that did not name it, none said "Machine Relations" (see [What AI Engines Call This Work](/research/what-do-ai-engines-call-ai-visibility-work-term-share-2026)).

## What this changes for measurement

- **Count citations and mentions separately.** A tracker that tallies brand names in the answer text would, on Perplexity in this panel, see about 8 percent of the sources the engine cited. One that tallies citation lists would credit sources the reader of the answer never sees named. Neither number stands in for the other. [Machine Relations](/glossary/machine-relations) treats citation and recommendation as separate measurements for this reason.
- **Read the engine before the rate.** ChatGPT's route names about half of what it cites and Perplexity's about one in twelve. A pooled figure hides that spread, and a move in the pooled figure can come from a change in which engine was sampled.
- **A citation is a retrieval event, not a reader event.** In this panel a cited source was named in the answer text less than one time in five. That is a statement about the answer, not about traffic, trust or any outcome for the source.

## What this page does not show

- **That the named source was the one the answer used.** A name in the answer text may come from an earlier part of the answer, from the model's own knowledge or from a different cited page, such as a list of tools in an answer that cites a comparison article. The count says the name appears. It does not link the name to the citation.
- **Recommendation.** A named source can be recommended, compared or warned against. Sentiment and position were not checked.
- **A cause.** Nothing here says a placement, a page change or a term caused any engine to name or cite anything.
- **Buyer behavior or consumer apps.** The runs are API routes: ChatGPT is mostly gpt-4.1-mini-2025-04-14 with 43 answers on gpt-4o-mini-search-preview-2025-03-11, Claude is claude-haiku-4-5-20251001, Perplexity is sonar, and Gemini is a Flash model. Signed-in consumer apps may name sources differently. The 49 questions are the Index's own panel and are not a sample of how buyers phrase them.
- **Names written another way.** The rule matches a brand name derived from the domain. Brands whose names differ from their domains, or which an answer abbreviates, are undercounted.
- **Change over time.** The panel was not sampled evenly across months.

**Method note.** Release mri_score_v2.1+2026-11-06+4a9cac5e2d8d; the [release manifest](https://machinerelations.ai/data/mri-release-manifest.json) records the artifact hash. Source: the Index answer ledger's run and citation-event tables, daily export of 2026-11-06 08:10 UTC, read with DuckDB and Node. Runs: success, answer text over 200 characters, category ai-visibility-geo, run date 2026-08-07 or later, engines ChatGPT, Claude, Perplexity and Gemini (1,129). Pairs: distinct root domains per run from the citation events (7,807), less 372 with a stem under five letters, 274 from general platforms (google, github, medium, wikipedia, forbes, amazon, microsoft, openai, anthropic, reddit, youtube, linkedin, quora, facebook, twitter, apple, nature, substack, notion) and 63 where the question contains the stem, leaving 7,098. Answer text was lower-cased, links removed with the pattern https?://, and tokens of the form name.tld with a common top-level domain removed before matching. Stem: first label of the root domain with non-alphanumerics removed, "try" or "get" dropped when six or more letters remain. The match is a whole word with an optional space between any two letters. A question contains the source when its letters and digits, joined, contain the stem. Intervals: question-resampled, 2,000 resamples, fixed seed. Answers without a stored citation list are not in the denominator.

## Frequently asked questions

### How often do AI engines name the sources they cite?

About one time in six, in this panel. Across 7,098 cited source-and-answer pairs from 899 answers to 49 AI-visibility buying questions, the answer text named the source in 17.6 percent. ChatGPT named 49.9 percent of the sources it cited, Gemini 17.1, Claude 13.1 and Perplexity 8.0.

### Does being cited by an AI engine mean the engine names your brand?

Not in these answers. In 466 of 899 answers none of the cited sources was named. Trackers and platforms the question was about were named often (Profound in 69 of 100 answers that cited it), while TechRadar, Search Engine Land and arXiv, cited in 133 answers between them, were named in 3.

### Which engine names the sources it cites most?

ChatGPT, at 49.9 percent, about 2.3 named sources per answer. Perplexity cites the most (11.5 per answer) and names the fewest, 0.9 per answer. Claude and Gemini fall between at 13.1 and 17.1 percent.

### What does this change about measuring AI visibility?

It means a citation count and a mention count are different measurements and the engine matters to both. A tool that counts names will miss most of what Perplexity cites; a tool that counts citation lists will credit sources the answer never names.

AuthorityTech, named in this report, is a related party to this site: it practices Machine Relations and does not publish machinerelations.ai, as disclosed at https://machinerelations.ai/about.

## Attribution

This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about

## Machine-readable related links

### Related concepts

- [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index)
- [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations)
- [MRI Score](https://machinerelations.ai/glossary/mri-score)
- [AI Visibility](https://machinerelations.ai/glossary/ai-visibility)

### Supporting research

- [API Answers Match Cited Text: Claude 84%, ChatGPT 37%](https://machinerelations.ai/research/where-ai-engines-copy-text-from-cited-pages-2026)
- [Do AI Citation Leaderboards Change Between Daily Releases? What the Machine Relations Index Re-Measures](https://machinerelations.ai/research/ai-citation-leaderboard-movement-between-releases-2026)
- [ChatGPT (API) Cited 18 of 25 Top Domains Blocking GPTBot and 4 of 11 Blocking OAI-SearchBot](https://machinerelations.ai/research/robots-txt-disallow-ai-citation-head-index-2026)
- [Google Search Console AI Performance Reports: What They Measure, What They Miss, and Why It Matters for Citation Architecture](https://machinerelations.ai/research/google-search-console-ai-performance-reports-citation-architecture)

### Framework context

- [Machine Relations Index](https://machinerelations.ai/index)
- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)