Claude’s web search returned no Reddit page in 245 answers to our own category’s questions, across five weekly runs in October 2026, where Google AI Mode cited Reddit in 31% of its answers to the same 50 questions.1 That runs against the assumption most AI visibility plans are built on: that the Reddit threads moving one engine move all of them.
The sample is small and it is ours: 50 questions about B2B marketing and AI visibility, asked five times. If your plan was built for ChatGPT and Google AI Mode and somebody has asked whether it covers Claude, these are the figures on what Claude cites when a buyer asks it to recommend software.
Empact Partners gets B2B software companies named when their buyers ask AI engines about the category. We run that as Generative Engine Optimization (GEO), and we built the Empact Panel to measure it every week across ChatGPT, Google AI Mode and Claude. Two runs of our own sit here beside the published studies and the records of Qvery, our sister company.
Highlights
Read together, they say Claude does not need a program of its own. It needs a different mix inside the GEO roadmap a team already runs: the Reddit half does nothing there, the lists other people publish do most of the work, and the traffic Claude sends sets how much of the budget that mix deserves.
Claude Has Not Cited Reddit in Any Sample We Hold
The Empact Panel asks ChatGPT and Google AI Mode 500 questions about our own category every week, and Claude a fixed 50 of them. For Claude it records every page its web search returned, which is a wider net than the pages an answer goes on to cite, so a zero there is a strong zero. Five runs, 4 to 30 October 2026:
Five runs and 245 answers, and the Reddit column in Claude’s row stayed empty every week, which makes it the steadiest number on the dashboard. On our category’s how-it-works questions Claude’s search reached for Wikipedia instead, in 16.3% of answers, where ChatGPT cited it in none.1
Our category is not yours, so we asked again in a software buyer’s words. On 4 November 2026 we put 120 buying questions (“What is the best CRM for a 50-person B2B sales team?” is the first) to ChatGPT, Google AI Mode, Gemini and Claude on the same day, and OtterlyAI had already counted a month of Claude’s citations in June:
Qvery is the other half of this measurement, at a scale we could not collect: it keeps the answers ChatGPT and Google AI Mode give to the buyer questions it follows, 724,019 of them since October 2025, and never asks Claude.6 In its late-September read across ten industries, Google AI Mode cited Reddit in 18.45% of buyer answers and ChatGPT in 0.89%.7
Qvery is our sister company, so discount its figures the way you would any vendor grading its own market, and read the method on each linked article before you lean on a number.
Three instruments, three kinds of question, and the same zero every time. We run Reddit work for partners and plan it for Google AI Mode, where the threads earn a place in the answers. That is the UGC half of our GEO work, built on what real users say about a product in public. In Claude it has nothing to earn, so a Claude plan inherits none of that budget.
.png)
Seven in Ten Pages Claude Cites Are Comparison Pages
On the 120 software questions we kept two layers of every Claude answer: the pages its search returned, and the pages its answer text cites. The second is the stricter one, and it is the one these figures use.
Prefer’s independent index ran 47 buyer questions through four engines in July and sorted every citation by the kind of site. Claude’s came out at 47% listicles, 36% vendors’ own sites and 15% media.8 Lists first is the shape our run found too, on a different question set four months later.
Many of these lists sit on software companies’ own blogs, Salesflare’s list of HubSpot alternatives among the ones Claude cited. OtterlyAI counted 64.0% of Claude’s June citations on company and product websites.5 Our run tested something narrower, whether the company that published each page was named in the same answer:
ChatGPT reads a vendor’s own page about the product it is recommending. Claude reads a list a company wrote about its category, and names somebody else from it nine times in ten.
So in Claude the mentions half of GEO, what independent pages say about you, is nearly the whole job. We do it with Existing Article Outreach, which places a partner’s product in the software lists already ranking for its category and quoted by the engines, as part of our Search Everywhere Optimization work.
A partner’s own comparison page still earns its keep in ChatGPT, where the publisher gets named more than half the time. Writing it is a Content Marketing job, and for Claude it is the smaller lever.
.png)
Claude Cites Review Directories Less Often Than the Other Three Engines
Review directories are where most software plans go next, and here Claude is the odd one out in the other direction.
Claude’s answers drop directory pages its own search found. Comparison pages hold: every answer whose search found one cited at least one.
Qvery’s October read of software buyers’ questions sets the two engines it tracks on a different question set: review sites sat behind 24.17% of ChatGPT’s answers to list questions and 18.61% of Google AI Mode’s.9 The profiles still feed Claude, one step removed, because the people writing the lists read them before they write.
ChatGPT’s Source List Will Not Stand In for Claude’s
A team that already tracks ChatGPT’s sources will want to reuse that list for Claude. On the same question, the two share too little to reuse, and three studies measured it three ways.
| Study | What was compared | What the two engines shared |
|---|---|---|
| Our software run | Claude and ChatGPT, the same question, 119 questions, 4 November 2026 | 5.8% of domains and 3.7% of URLs2 |
| OtterlyAI | Claude and ChatGPT, a month of citations, June 2026 | 13% of domains and 4.2% of URLs5 |
| The Prefer Index | Four engines, 47 questions, July 2026 | 76% of 713 domains were cited by one engine only8 |
The answer holds across all three. What ChatGPT cites tells you little about what Claude will cite, and in our run the gap reached down to single questions:
So Claude gets read on its own answers. The engine it resembles is Gemini, which shared more than twice the domains with Claude that ChatGPT did and leaned on comparison pages almost as heavily, which is why a list placement earned for one tends to serve the other.
Claude Searches Once, Through Brave, and Cites From That List
Anthropic lists Brave Search as the subprocessor behind web search in every Claude product.10 The index Claude shops in is Brave’s.
To see whether that shows in the answers, we sent the same 120 questions, as typed, to Brave’s search and to Google’s on the day of the run, and kept the top 20 results of each:
Claude’s search reads Brave’s results for something close to the buyer’s own words, keeps a short list and quotes most of it. Brave had a Reddit page for most of these questions and Claude’s search never passed one on, so something between Brave’s index and Claude’s answer drops Reddit, and we cannot see what.
What a team can act on is the rest. The lists that rank in Brave for the buyer’s own wording are Claude’s candidate list, and they are not the lists Google ranks.
Qvery’s mid-October read against Google’s first page found that 89.6% of the sites ChatGPT cited sat off Google’s first page for the same question, and 76.8% of Google AI Mode’s.11 A Google rank report was never the map of AI citations, and for Claude it reads the wrong index.
Claude Sends Little Traffic and Keeps a Page for Longer
Whether Claude earns a budget line depends on how many visits it sends and how long a citation there lasts.
Profound followed 883,000 pages across seven engines for a year. Among pages that last beyond two weeks, a Claude citation was 2.8 times as likely as a ChatGPT one to still be relevant eight weeks later, and 63% of Claude’s lasting pages were cited across weeks before their peak, against 18% to 26% on five of the other six engines.12
We have no partner figure for the traffic Claude sends, because no public case study of ours publishes one, so the arithmetic here is Statcounter’s and Profound’s rather than ours.
Size the Claude line by its traffic, which is small. Spend it on the lists, where one placement also serves Gemini and, on Profound’s count, lasts longer than it would in ChatGPT.
What Changes in a GEO Roadmap When Claude Is on It
When we add Claude to a partner’s roadmap, the change lands in four places, and none of them is a new program.
That sits inside the GEO work we run for B2B software companies: an audit of the questions a partner’s buyers ask and where each engine names them, then a roadmap that sequences the lists with the SEO and content work. A named senior consultant runs it, specialists join for each workstream, and the Empact Panel ships the standing reports, engine by engine.
The horizon is quarters, because the lists change slowly and GEO takes months before the answers move. So watch one number over the next quarter: the share of your category’s Claude answers that cite a list your product is in. If it rises while your Reddit figures stay flat, the mix is right.
If your AI visibility plan was built for ChatGPT and Google AI Mode and someone has asked whether it covers Claude, book a call with me, and let’s see whether we could work together.
Sources
- Empact Partners, the Empact Panel, our own category: 500 questions on B2B marketing services and AI visibility for ChatGPT and Google AI Mode, a fixed 50 of them for Claude (for Claude, the pages its web search returned), five runs on 4, 9, 18, 23 and 30 October 2026, 5,250 answers; read 4 November 2026.
- Empact Partners, our software run: 120 B2B software buying questions in 15 categories, US English, one answer each from ChatGPT with web search, Google AI Mode, Gemini and Claude (Sonnet 4.6 with Anthropic’s web search tool; the pages its answer cited and the pages its search returned), 4 November 2026.
- Empact Partners, our search-results check: the same 120 questions as typed, sent to Brave’s Web Search API and to Google (through DataForSEO), top 20 results each, United States, English, 4 November 2026.
- Statcounter Global Stats, AI Chatbot Market Share Worldwide, monthly, April 2025 to September 2026. Independent: referrals from AI chatbots to over 1 million websites; downloaded 4 November 2026.
- OtterlyAI, Claude AI Citations Study: How to Get Cited in 2026, 2026. Vendor study: 379,321 Claude citations across 16,406 domains, June 2026.
- Qvery, our sister company, the GEO Agent’s records: 724,019 completed answers, ChatGPT since 28 October 2025 and Google AI Mode since 18 December 2025, no other engine; read 4 November 2026.
- Qvery, our sister company, Reddit AI Citation Statistics (2026), 2026. Buyer questions in ten industries in the US, UK, Canada and Australia, late September 2026.
- Prefer, The Prefer Index, Edition 01: Who Gets Cited in AI Search, 2026. Vendor study: 47 buyer questions, four engines, 1,237 citations, run 10 July 2026.
- Qvery, our sister company, G2 and Review-Site AI Citation Statistics (2026), 2026. Software buyers’ questions on ChatGPT and Google AI Mode, October 2026.
- Anthropic, Trust Center: Subprocessors, the Brave Search entry (web search, all products); read 4 November 2026.
- Qvery, our sister company, AI Citations vs Google Rankings Statistics (2026), 2026. US buyer questions on ChatGPT, Google AI Mode and Google, 15 October 2026.
- Profound, The half-life of an AI citation is 11 days, 2026. Vendor study: 883,000 pages, 1.19 million page and engine lifecycles, seven engines, September 2025 to September 2026.
