Claude SEO: how to get your brand recommended by Claude
Claude SEO is getting your brand named, and your pages cited, when people ask Claude, Anthropic's assistant, what to buy. Claude answers from what it learned in training, and from the web when web search is on. So you need Anthropic's search crawlers allowed and a reputation on the pages it learns from.
Picked is an AI visibility tool for SEO and marketing agencies, and B2B SaaS teams. Every day it asks ChatGPT, Claude, Gemini, Perplexity and Google's AI answers the prompts your buyers ask, records which brands and pages they name, and writes the article built to become the source they quote.
Here to use AI for SEO work? Picked also finds your keyword gaps and writes the article.
3
Anthropic crawlers, each its own robots.txt rule
2
of them, Claude-SearchBot and Claude-User, decide today's search answers
From Anthropic's crawler article, updated April 7, 2026.
How it picks
How does Claude decide which brands to recommend?
Claude has two sources for a buying question.
Training
Claude was trained on a large crawl of the web, collected by Anthropic's
ClaudeBot, among other sources. Brands that were well described on many pages before its cutoff can be named from memory.Web search
When web search is on, Claude searches when "you ask about topics that benefit from current information", in the words of Anthropic's help page, and every answer that searches includes citations to the pages it used.
The difference from ChatGPT and Perplexity is how often the second source is used. Anthropic's help page tells people to switch web search on from the "+" menu, and Team and Enterprise owners have to enable it first. Anyone who hasn't switched it on gets answers written from training alone, with no search and no links. For those, what the web said about you before Claude's cutoff is all that counts.
So Claude SEO has two jobs, on two clocks:
What decides it
- Answers with web search
- The pages Claude finds and reads today
- Answers from training
- What the web said about you before the model's cutoff
How fast it moves
- Answers with web search
- Days to weeks
- Answers from training
- Only when Anthropic ships a new model
What you change
- Answers with web search
- Crawler access, pages that answer the question
- Answers from training
- Being described, consistently, on many pages
Shows a link to you
- Answers with web search
- Yes, as a citation
- Answers from training
- No
| Answers with web search | Answers from training | |
|---|---|---|
| What decides it | The pages Claude finds and reads today | What the web said about you before the model's cutoff |
| How fast it moves | Days to weeks | Only when Anthropic ships a new model |
| What you change | Crawler access, pages that answer the question | Being described, consistently, on many pages |
| Shows a link to you | Yes, as a citation | No |
The crawlers
Which Anthropic crawlers matter?
Anthropic runs three, all controlled by robots.txt. From Anthropic's crawler article, updated April 7, 2026:
ClaudeBot- What it does
- Collects pages that could be used to train Anthropic's models
- Block it and
- Your future pages are left out of training
Claude-SearchBot- What it does
- Indexes pages to improve Claude's search results
- Block it and
- Anthropic says it "may reduce your site's visibility" in search answers
Claude-User- What it does
- Opens a page when a person asks Claude a question
- Block it and
- Claude can't fetch your page to answer that person
| Crawler | What it does | Block it and |
|---|---|---|
ClaudeBot | Collects pages that could be used to train Anthropic's models | Your future pages are left out of training |
Claude-SearchBot | Indexes pages to improve Claude's search results | Anthropic says it "may reduce your site's visibility" in search answers |
Claude-User | Opens a page when a person asks Claude a question | Claude can't fetch your page to answer that person |
Blocking ClaudeBot is the one with a slow cost: it doesn't change today's answers, but the next Claude model learns about your category from everyone else's pages. If you want to be named from memory in a year, think twice before blocking it. Claude-SearchBot and Claude-User affect answers now.
A Claude-User hit in your logs is a person asking Claude a question and Claude opening your page to answer it. A ClaudeBot hit is a crawl for a training set, someday. Only one of those is a buyer.
Track the real app, not the API, where you can.
What to change
How do I get my brand recommended by Claude?
No one can make Claude say your name. These are the steps that give it reasons to.
- 1
Allow
Claude-SearchBotandClaude-UserCheck
robots.txt, your CDN's AI bot settings and any firewall rule that challenges bots. Decide aboutClaudeBoton purpose, not by a one-click "block AI" switch. And keep the facts in the page's HTML: Anthropic's crawler fetches JavaScript files but doesn't run them, according to Vercel's crawler data (December 2024). - 2
Be described the same way everywhere
For answers from training, Claude repeats what many pages agreed on. Your category, who you're for and what you cost should read the same on your site, directories, review sites and comparison pages. Five pages calling you five different things teach a model nothing.
- 3
Get onto the pages that name the options
"Best X for Y" roundups, review sites, directories, forums where buyers ask. These feed both roads: they're in the training crawl, and they're what web search finds.
- 4
Answer each buying question on its own page
Pricing, comparisons with named competitors, "best X for [use case]". First sentence answers the question in words Claude can quote.
- 5
Write facts that can be checked
Prices, limits, dates, integrations, who it isn't for. Specifics survive being summarized; slogans don't.
- 6
Keep doing SEO
When Claude searches, it finds pages through a search index. Pages that are indexed and linked get found.
What doesn't help: pages written "for Claude" in a special format, hidden prompts aimed at AI, or an llms.txt file as the whole plan. Anthropic's crawler article says nothing about llms.txt.
Measure it
How do I check if Claude recommends my brand?
Ask it, twice: once with web search on, once with it off. The first shows what Claude finds today, with links you can open. The second shows what it learned in training. If you're in the first and not the second, the pages are working and the next model may catch up. If you're in neither, start with steps 2 and 3.
Then check your logs for Claude-User and Claude-SearchBot. No hits at all usually means something is blocking them. Visits from Claude's links are easier to count than most: Ahrefs' referrer tests (May 2025) found Claude passed the referrer on every link they tried, so it shows up as claude.ai in analytics.
To track it over time, an AI visibility tool asks the same prompts on a schedule. Picked asks Claude less often than the others, since a Claude answer costs us several times more than the other models. There's no public way to read the answer the Claude app shows a signed-in person, so Picked asks through Anthropic's API with web search on, and says so next to the number. Tools that pass API answers off as "the Claude app" are measuring something your buyers never see. The same goes for the other AI models: where the real product's answer can be read, Picked reads that.
LaunchOnIt in Picked: 5 prompts, named in none
When not to bother
When Claude SEO won't pay off
Honest answer: sometimes this isn't worth your time.
Your buyers don't use Claude
Plenty don't. Check before spending time on it; ChatGPT or Google's AI answers may matter more.
You're new
A brand launched after Claude's training cutoff can only be found through web search until the next model. Build the pages and listings now, and expect the memory side to lag.
You want a guarantee
Nobody controls what Claude says. Anyone promising a recommendation is guessing.
Sources
How we checked
- Anthropic: the crawler article (updated April 7, 2026) and the web search help page, both read September 29, 2026.
- Third-party studies: Vercel's AI crawler data (December 2024), Ahrefs' referrer tests (May 2025), read September 30, 2026.
- Picked: how Claude is asked is from our own setup: Anthropic's API with web search, US, less often than daily per prompt. We don't have enough stored Claude answers yet to publish citation figures, so this page has none.
The other AI models
Each one reads the web its own way
Questions
Frequently asked questions
Short answers, each one checked against the vendor's own docs or our own answers.
Yes, when web search is on. Anthropic's help page has people switch it on from the "+" menu, and Team and Enterprise owners enable it first. With it on, Claude searches when a question needs current information, and its answer includes citations to the pages it read.
Yes. ClaudeBot collects pages for training; Claude-SearchBot and Claude-User are what search answers rely on. Blocking ClaudeBot keeps your future pages out of training, so Claude's next models learn less about you, but today's search answers can still cite you.
Claude-User is the agent Claude uses to open a web page when a person asks it a question. A Claude-User hit in your server logs means someone asked Claude something just now and it read your page to answer. Blocking it stops Claude fetching your page for those people.
They're trained on different data, search differently and write answers differently. Across the models we track, most brands named for a question are named by only one of them. Track each model your buyers use separately rather than assuming a win on one carries to another.
Yes, and that's what most searches for Claude SEO are about: Claude skills, plugins and Claude Code workflows for audits, keyword research and writing. It's a different job from getting Claude to recommend your brand, which is what this page covers.
Find out what ChatGPT and Google AI say about you
Add your domain and the prompts your buyers ask. Unlimited seats for your whole team, and no promises we can't keep.