Guides · Updated October 19, 2026
How to Get Your Brand Recommended by Claude
See where you stand first. The free check asks ChatGPT, Perplexity, Gemini and Claude five buyer questions about your category and shows who they name and which pages they cite. No account needed.
Run the free checkClaude is easy to overlook when you think about AI search, but plenty of people ask it what to buy. It works in two modes. Sometimes it answers from what it learned in training. Sometimes it searches the web first and cites what it finds. To be recommended, you need a plan for both.
The short version: let Anthropic’s search crawlers in, don’t hide pages from search with noindex, get named on the pages Claude cites for your buyers’ questions, and keep measuring.
How Claude builds an answer
When a question needs current information, Claude calls a web search tool, reads the results and writes an answer with citations to the sources it used. Anthropic says every web-search response includes citations (Claude Help Center). In Claude’s API, Anthropic describes the same loop: Claude decides when to search based on the prompt, the API runs the search, and Claude replies with cited sources (Claude API docs).
Three things follow from what Anthropic documents:
- Search isn’t always on. Claude decides when to search. In the Claude apps, web search can be switched on or off per chat (on Team and Enterprise plans an owner has to enable it first), and in the newer Claude experience there’s no toggle and Claude searches when it helps (Claude Help Center).
- Claude’s web results come from search partners. Anthropic says a
noindextag tells its partners not to index your content, so they don’t send it to Claude, and the page won’t appear in Claude outputs that use web search (Claude Help Center). Anthropic doesn’t name the text search partner. It does say image results are powered by Bing. - Location can matter. Anthropic notes Claude may use a user’s location, inferred from their IP address, for localized results (Claude Help Center), so a local business may see different answers than a buyer in another country.
From our own checks: across 160 Claude answers our checker collected up to October 19, 2026 (28 checks, mostly our own test runs on common software categories), Claude ran a web search in 124 (78%). The rest were answered from memory. Our setup uses Claude Haiku 4.5 through the API with one search allowed per answer, so the Claude apps may search more or less often. Either way, a real share of answers never touch the live web.
That’s why this guide covers both modes: search answers depend on what’s indexed and ranks for the question today, and memory answers depend on how widely and consistently your brand is described across the web over time.
Step 1: See what Claude says about you today
- Write 5–10 questions a real buyer would ask, without your brand name: “best X for Y,” “alternatives to [category leader],” “X vs Y,” and one about a specific use case.
- Ask each one in a new chat, with web search available, two or three times. Answers vary between runs.
- For each answer, note: did Claude search, were you named, in what position, how you were described, which competitors appeared, and every cited URL.
Split your notes by whether Claude searched. If you’re missing from searched answers, the fix is mostly about pages (Steps 2–4). If you’re missing from answers Claude gave without searching, the fix is slower and is about being mentioned widely (Step 5).
Step 2: Let Anthropic’s crawlers in, and don’t noindex key pages
Anthropic uses three separate user agents, each controlled in robots.txt (Claude Help Center):
- Claude-SearchBot analyzes content to improve the relevance and accuracy of search results. Anthropic says blocking it may reduce your visibility in user search results.
- Claude-User fetches a page when someone asks Claude a question that needs it. Anthropic says blocking it prevents Claude from retrieving your content for that user, which may reduce your visibility in user-directed web search.
- ClaudeBot collects content that could contribute to training. Blocking it tells Anthropic to leave your future content out of training data. That’s a separate decision from search.
What to check:
- robots.txt. Open
yourdomain.com/robots.txt. Make sure no rule disallowsClaude-SearchBotorClaude-User, either by name or through a blanketUser-agent: *rule. Anthropic asks you to set rules on every subdomain you care about. - Firewall or CDN bot rules. Anthropic says blocking its IP addresses isn’t a reliable way to control its bots, because it stops them reading your robots.txt. The reverse also applies: an aggressive “block AI bots” switch at your host can shut out Claude-SearchBot even when robots.txt allows it.
- noindex. Your homepage, pricing, features and comparison pages should not carry a
noindextag. Anthropic says noindex keeps a page out of Claude’s web search answers (Claude Help Center). Staging-site tags that ship to production by accident are a common cause. - Plain HTML. What you do, who it’s for, pricing and key features should be readable text, not only rendered by JavaScript or shown in images.
Step 3: Get named on the pages Claude cites
Go back to the cited URLs from Step 1. Mark the ones that name competitors and not you. Those are your targets.
- “Best X” roundups: send the author facts they can check: what you do differently, who it’s for, current pricing, and a free account to test. An update to a page Claude already cites helps more than a new page it never finds.
- App stores, review sites and directories that show up in your citations: complete your listing so it says plainly what the product does and who it’s for, and ask real customers for detailed reviews. Never write or buy reviews.
- Community threads (Reddit, forums, Indie Hackers): answer honestly under your own name and say you built the product.
- Outdated descriptions: if a cited page has old pricing or a removed feature, ask the publisher to fix it. Claude can only repeat what the page says.
Step 4: Make your own pages easy to cite
- Use the buyer’s question as a heading and answer it in the first sentence or two underneath.
- Make each section stand on its own. A quoted paragraph should still say what the product is and who it’s for.
- Publish honest comparison pages (“[You] vs [Competitor]”) that say when the other product is the better fit. For “X vs Y” questions, a fair page on your own site is often the best source available.
- Keep key pages current and dated. In Claude’s API, each search result includes when the page was last updated (Claude API docs), so a visible, accurate date on pricing and feature pages is worth keeping current.
- Add evidence, not keywords. In the GEO study from Princeton and IIT Delhi researchers (KDD 2024), adding citations, quotations and statistics made sources more visible in generative engine answers, while keyword stuffing did little (Aggarwal et al.). It’s a research result, not a guarantee, and the study didn’t test Claude specifically.
Step 5: Be mentioned widely, for the answers Claude gives from memory
When Claude answers without searching, it relies on what it learned in training. Anthropic doesn’t publish how that shapes brand recommendations, and you can’t submit your brand to it. What tends to help any model know a product is the same work that helps search: being described consistently, in the same words, on many independent pages (your site, directories, reviews, articles, community threads), over months, not days.
Two practical rules:
- Use one clear one-liner everywhere. Same category words, same audience, on your site and every listing.
- Don’t expect fast changes here. Search answers can change as soon as a page is indexed and ranks. Memory answers only change when a newer model is trained.
Step 6: Measure on a schedule
Re-run the same questions weekly or monthly and compare:
- The share of answers that name you, split by searched vs not searched
- Your position, and whether the description is accurate
- Which competitors appear
- Which domains Claude cites, and whether your target pages have started naming you
Look for changes that hold across several questions and runs, not one answer. Expect different results on other assistants: our guides on ChatGPT and Perplexity cover how their crawlers and citations differ.
If you’d rather not run this by hand, WhatTheBotSaid’s free check asks Claude, ChatGPT, Perplexity and Gemini five buyer questions about your category and shows who they named, which pages they cited, and whether you made the list. Paid plans re-run the same questions every week so you can see the trend.
Common mistakes
- Blocking all of Anthropic’s bots when you only meant to opt out of training. ClaudeBot, Claude-SearchBot and Claude-User are separate rules.
- A leftover noindex tag on pages you want buyers to find.
- Testing with your brand name in the question. “Is [Brand] good?” doesn’t tell you whether Claude recommends you to someone who hasn’t heard of you.
- Judging from one answer. Claude doesn’t search every time, and answers vary between runs.
- Astroturfing forums or review sites. It gets removed and can cost you your listings.
FAQ
How does Claude decide which brands to recommend?
Claude answers either from what it learned in training or, when web search is on and the question needs current information, from a live search. With search, it reads the results and writes an answer with citations, so the brands it recommends are mostly the ones named on the pages it found. Anthropic doesn’t publish how it ranks sources or picks brands.
Does Claude search the web for every answer?
No. Anthropic says Claude decides when to search based on the question, and in the Claude apps web search can be switched on or off. In our own checks, Claude used search in about three out of four answers when the tool was available. The rest came from the model’s own knowledge.
Which crawler does Claude use, and should I allow it?
Anthropic runs three: ClaudeBot (collects content that may be used for training), Claude-User (fetches a page when a user’s question needs it) and Claude-SearchBot (indexes content to improve search results). Anthropic says blocking Claude-User or Claude-SearchBot may reduce your visibility in Claude’s search answers. Each one is controlled separately in robots.txt.
If I block ClaudeBot, will Claude stop recommending me?
Blocking ClaudeBot tells Anthropic to leave your site’s future content out of training data. It is a separate rule from Claude-SearchBot and Claude-User, which are the ones tied to search answers. Decide on each one separately.
Can I pay or submit my site to be recommended by Claude?
Anthropic doesn’t document any way to pay for placement or submit a site for Claude’s answers. What it does document is crawler control through robots.txt, and that a noindex tag keeps a page out of Claude’s web search results.
How do I check whether Claude recommends my brand?
Ask Claude a fixed set of buyer questions in new chats, with web search on, a few times each, and note who it names and which sources it cites. You can do it by hand, or run a free check at whatthebotsaid.com to see Claude’s answers next to ChatGPT, Perplexity and Gemini.
Sources
- Claude Help Center, “Does Anthropic crawl data from the web, and how can site owners block the crawler?”
- Claude Help Center, “Report, block, and remove content from Claude”
- Claude Help Center, “Enable and use web search”
- Claude API docs, “Web search tool”
- Aggarwal et al., “GEO: Generative Engine Optimization,” KDD 2024
- WhatTheBotSaid check data, 160 Claude answers across 28 checks, as of October 19, 2026