- Claude as a tool: excellent for analysing Search Console exports, drafting, technical review and building audit workflows. It ships nothing by itself and knows nothing about your site until you feed it.
- Claude Code has a real SEO skill ecosystem, including open-source toolkits running dozens of sub-skills and specialist agents against a site.
- Claude as a surface: in Profound's July 2026 study, Claude searched the web in 93% of sampled responses while Claude Code searched in just 13%.
- The same study found the two surfaces mentioned only about 20% of the same brands on average for the same prompt, and fetched very different page types.
- Practical takeaway: record the product and interface, not just the model name, in any AI visibility measurement.
Claude does two jobs in SEO, and they need separating. As a tool, it is strong at analysis, drafting and auditing when you connect it to real data, and there is now an ecosystem of open-source SEO skills built for Claude Code. As a search surface, Claude is somewhere your buyers ask for recommendations, and published research puts Claude and Claude Code closer to two different systems than one: one searches the web in almost every response, the other rarely does, and they name mostly different brands for the same question.
Can Claude do SEO?
Yes, for the analysis and drafting half of the job. Feed it a Search Console export and it will find queries with high impressions and no clicks, cluster topics, spot cannibalisation and draft the fixes. Practitioners describe running whole workflows this way, and the open-source tooling has caught up: projects like the Claude SEO skill for Claude Code crawl a site and produce scored SEO, GEO and AEO audit reports, coordinating multiple specialist sub-skills in one run.
What Claude cannot do is the part that actually moves rankings: deciding what ships, changing the live site, and verifying the change landed. Nothing in the model knows your margins, your commercial priorities or whether a page's claim is still true. The boundary between what to automate and what to keep human is set out in where the judgement boundary sits, and the permission mechanics in giving a system write access safely. Both apply directly if you point an agentic toolkit at a production site.
The part almost nobody covers: Claude and Claude Code are different answer engines
If you also care about being recommended by Claude, treat the surface with the same precision you would give a search engine. Profound published research in August 2026 comparing how Claude and Claude Code respond, and the numbers are worth reading carefully, with the caveat that they are one vendor's first-party study.
The sample: 24,135 responses across 1,724 prompts, with a coding subset of 2,800 responses across 200 prompts, collected 13 to 23 July 2026. The findings Profound reports:
- Search behaviour: Claude searched the web in 93% of sampled responses; Claude Code searched in 13%.
- Brand overlap: for the same prompt, the two surfaces mentioned only about 20% of the same brands on average.
- Response shape: Claude averaged 459 words, Claude Code 322. Lists appeared in 56% of Claude responses against 94% for Claude Code; tables in 11% against 54%.
- Pages fetched: across the top 1,000 pages each agent visited between 18 July and 18 August, Profound reports nearly three-quarters of Claude Code's visits went to documentation, informational and pricing pages, while around 60% of Claude's went to robots.txt files, sitemaps and home pages.
Read those as directional, not settled. The figures come from Profound's own panel and page-type classification, they cover a ten-day prompt window, and they will move with prompt set, geography, model version and site sample. What survives the caveats is the structural point: a product built on the same underlying model can retrieve differently, cite differently and recommend differently. Anyone reporting "Claude visibility" as one number is averaging two different systems.
What that means for your pages
If a coding-context surface leans on documentation and pricing pages, then for technical products those pages are commercial surfaces, not afterthoughts. Practical implications, none of which require exotic tactics:
- Put the facts that decide a recommendation (what it does, what it costs, what it supports, what it does not) in readable HTML on stable URLs.
- Keep documentation and pricing pages crawlable, current and specific. Vague pricing pages lose recommendations they never knew they were in.
- Do not treat robots.txt and sitemaps as neutral plumbing. If agents fetch them constantly, they are part of your discoverability.
- Record which surface an observation came from. Model, product, interface, prompt, market and date, every time.
That last one is the measurement discipline in AI search metrics: what to measure, and it is why a single blended "AI visibility score" is weak evidence for a build decision.
Is Claude better than ChatGPT for SEO?
For long-document analysis and code-adjacent work, many practitioners prefer Claude. For breadth of plugins and integrations, ChatGPT has a larger ecosystem. Honestly, the tool preference matters less than whether real data goes in and a person reviews what comes out. The bigger difference is on the visibility side: they are separate surfaces with separate retrieval, so being cited in one tells you nothing about the other.
How we use this at Searchmaxxed
Agents do the reading, drafting and verification inside our Managed Search Loop, and senior operators own every decision to ship. On the visibility side we keep per-surface observation records rather than one blended score, and the fixes land as page and source work through the Agentic Website. If you want to know where you currently stand across Google and AI answers, the AI Visibility Audit maps it in six business days for A$4,500, credited in full if a program starts within 14 days.
FAQs
Can Claude do SEO?
It can analyse data, draft content and run audit workflows, especially through Claude Code with SEO skills installed. It cannot decide what ships, change your site safely without guardrails, or verify results. Treat it as a fast analyst, not an operator.
What is the Claude SEO skill?
Open-source SEO toolkits built for Claude Code that crawl a site and produce SEO, GEO and AEO audit reports, coordinating multiple sub-skills and specialist agents in a single run. They are free and genuinely useful for detection work.
Which AI is best for SEO?
The one connected to your real data with a human review step. Model choice matters far less than whether Search Console, analytics and your actual pages are in the loop.
Is Claude better than ChatGPT for SEO?
Claude is often preferred for long-document analysis and code work; ChatGPT has broader integrations. For visibility purposes they are different surfaces entirely and should be measured separately.
Do Claude and Claude Code give the same answers?
Profound's July 2026 study found they mentioned only around 20% of the same brands on average for the same prompt, with Claude searching the web in 93% of sampled responses against Claude Code's 13%. Treat them as separate systems.
How do I get cited by Claude?
The same fundamentals that earn any citation: crawlable pages, specific and current facts in readable HTML, documentation and pricing that answer real questions, and third-party sources that corroborate you. There is no Claude-specific trick worth the risk.
Should I optimise my documentation for AI?
Optimise it for readers first and keep it crawlable and current. If coding-context agents lean on docs and pricing pages, as Profound's sample suggests, then good documentation is already commercial work.
Can I use AI to do all my SEO?
You can automate the reading, drafting and verification. Automating publishing at scale walks into Google's scaled content abuse policy, so keep a person on the decide-and-publish steps.
Are Profound's Claude figures reliable?
They are first-party vendor research with a stated sample and window, which puts them well above most numbers in this category, and still short of an independent benchmark. Use them to shape hypotheses and measure your own surfaces yourself.
Next step
Point Claude at your Search Console export this week and see what it finds. When the findings need shipping and verifying, that is our job: LET'S TALK.