Future-proof your SEO strategy by creating content your audience is actually searching for
SEO Simplified will help you identify the words your audience is typing into the search bar, so you can create content that connects with humans, not just search engines.
Subscribe to get your FREE Copy of the SEO SIMPLIFIED keyword research workbook.
Jump To:
Claude can do a lot for SEO. But some of the jobs we’re handing it are impossible from the start..
AI experts are selling Claude “SEO skills” that promise to audit your site, find your best keywords, and optimize your content.
At the same time, business owners are asking for things like, “I want a skill that will update my website for Answer Engine Optimization (AEO).”
I get the appeal. If Claude is already doing a ton of heavy lifting in your business, why not hand it some of your SEO too?
You absolutely can. But you have to give it the right job.
SEO is not one task, and Claude cannot do every SEO job just because a skill tells it to.
Sometimes it needs a much more precise definition of what you’re asking it to do. Other times, it needs data or access it simply doesn’t have.
I use Claude constantly in my own SEO work. But I give it very specific jobs, along with the frameworks, criteria, and data it needs to do them well.
So if you’re trying to figure out how to use Claude for SEO, the real question isn’t whether Claude can do it.
It’s what it can actually do well, what it needs from you, and where those SEO skills start to overpromise.
Why "optimize this page" is a terrible prompt
“Optimize this page” sounds like a clear instruction, but it leaves Claude to decide what “optimized” is supposed to mean.
Optimize for what?
Who’s the audience? What keyword are you targeting? What have you already written about elsewhere on your site? Are you optimizing for traditional Google search? Local SEO? AI search?
Those questions change what Claude should recommend.
If you don’t define the goal and give Claude the context it needs, it’ll fill in the blanks itself. And it’s very good at producing an answer that sounds thoughtful and specific, even when it’s solving a different problem than the one you actually have.
Claude needs criteria, not just a goal
Even once Claude knows what you’re trying to accomplish, it still needs to know how you want it to evaluate the page.
What does a strong result look like? What information has to be included? What should it leave alone? What matters most if there are tradeoffs?
Without those criteria, Claude still has to make some of those decisions for you.
That’s where things can get slippery. It might add keywords you never researched, recommend content you’ve already covered elsewhere, or improve one part of the page while creating a new problem somewhere else.
The more clearly you define the job and the rules for doing it, the better Claude gets at executing it.
This is why I use skills in my own SEO work
I use Claude for SEO a lot, but I don’t just hand it a URL and ask it to make things better.
I’ve spent a lot of time building skills that give Claude the definitions, rubrics, frameworks, and decision rules I want it to use.
So I’m not asking Claude to decide what a strong hero section looks like, what counts as a useful internal link, or how I want a page evaluated. I’ve already made those decisions and built them into the skill.
That’s a big part of why Claude can be so useful for implementation.
The less it has to improvise the strategy, the better it can execute the task.
Claude can only work with the information it has
Even a perfectly defined SEO task will fall apart if Claude doesn’t have access to the information it needs to do the job.
Claude is very good at analyzing, comparing, organizing, spotting patterns, applying a rubric, and turning a pile of information into something useful.
But it can’t analyze information it never received.
Claude becomes much more useful for SEO when you give it real data instead of asking it to infer what the data might say.
That might be Google Search Console data, keyword research, a website crawl, backlink data, a list of your existing content, or competitor research.
Give Claude that information and it can help you find patterns, compare opportunities, prioritize what matters, and turn a messy spreadsheet into a much more manageable list of next steps.
For example, I can give Claude Search Console data and ask it to identify queries that are within striking distance of the top three ranking spots, or pages that are losing clicks and impressions over time. I can also ask it to identify keywords that are driving traffic to my website but could be captured by a newer, better optimized piece of content.
That’s a very different job from asking Claude to guess what people are searching for in the first place.
What giving Claude a URL actually means
Claude can learn a lot from a webpage URL, but fetching a page is not the same thing as performing a a complete website audit.
WebFetch is the tool Claude uses to retrieve and read the content of a webpage from a URL you give it.
When Claude fetches a page, it can work with useful information such as the page copy, H1, H2, and H3 structure, links and anchor text, images represented in the HTML, alt text when it’s exposed, and image URLs and filenames in some cases.
That’s plenty for useful page-level SEO work.
What it doesn’t give Claude is a complete picture of every page on your website or every piece of information behind the page.
Claude can work with what WebFetch exposes
WebFetch gives Claude enough information to evaluate many important on-page SEO elements.
If I give Claude a service page and a specific search I want that page to support, it can help me evaluate whether the service is clearly explained, whether the page actually covers the topic, whether the headings make sense, whether important information is missing, and whether the copy supports the search I’m targeting.
It can also compare pages you give it, look for overlap, and evaluate the internal links it can see.
That’s exactly the kind of work Claude can be very good at.
The problem starts when we assume that because Claude can read a page, it can see everything about the page or everything across the site.
Claude can't inspect information WebFetch doesn't expose
Anything that falls outside what WebFetch provides is effectively invisible to Claude unless another tool or data source supplies it.
Images are a good example.
If an image is included directly in the HTML, Claude may be able to see its URL, filename, and alt text.
But if that same image is being used as a CSS background on a container, which is common for hero sections, WebFetch may not expose it at all.
To you, both versions look like an image sitting at the top of the page. To Claude, one may be visible and the other may not exist at all.
The same problem shows up at the site level. Claude might successfully fetch your homepage, service pages, and a handful of blog posts without having a complete inventory of every URL that exists on your site.
That means it can’t reliably tell you whether it found every page, whether you have orphan pages, or whether it has audited the entire site just because it followed a few links.
Claude can only audit what the workflow actually exposes to it.
Seeing a webpage is not the same as measuring it
Claude can do a lot with the right inputs. But there’s another distinction that matters: being able to see something isn’t the same as being able to measure it.
Claude can directly observe things like the text of your H1, the copy on the page, the anchor text of an internal link, or the filename of an image exposed in the HTML.
But image file size, impressions, clicks, rankings, page speed, and backlink counts require actual data.
Seeing an image doesn’t mean Claude knows its file size. Seeing a keyword on the page doesn’t mean Claude knows how many people search for it. Seeing a page doesn’t mean Claude knows what it currently ranks for.
That distinction sounds obvious once you say it out loud. But a lot of AI-generated SEO advice crosses that line without making it clear.
Why some Claude SEO skills overpromise
A Claude skill can give the model excellent SEO instructions, but it cannot automatically give Claude access to information the workflow does not provide.
This is where I see a lot of the overpromising happen.
A skill may have a detailed process for auditing a website or optimizing a page. That process may even be smart.
But there’s still a difference between what the skill promises, what Claude is instructed to do, and what information Claude can actually access.
If those three things don’t line up, the output can sound much more authoritative than it should.
Good instructions do not create missing data
A skill can give Claude perfectly good instructions for a job it still doesn’t have the information to complete.
Schema is a good example.
I could build an SEO audit skill that tells Claude exactly how to evaluate a webpage for the presence and accuracy of its schema markup. I could give it a detailed rubric explaining what schema should be there, what properties to check, and what problems to flag.
Those instructions could be excellent.
But WebFetch doesn’t preserve the schema markup for Claude to inspect.
So when the skill tells Claude to evaluate the schema on that URL, Claude doesn’t actually have the information it needs to do it. And the particularly tricky part is that it may simply move on rather than stopping to tell you, “I can’t see the schema.”
A detailed instruction can tell Claude what to look for. It can’t make the missing information appear.
A good skill should make its limits clear
A well-designed SEO skill should know when it doesn’t have enough information to complete the task.
If the skill is supposed to check something Claude can’t actually access, the right response is to stop and tell you what’s missing, not quietly skip the check or fill in the blank.
A skill that says, “I can’t determine that from the information available. Give me the data I need and I can analyze it,” is doing a better job than one that confidently invents an answer.
How to tell whether you should trust a Claude SEO recommendation
The easiest way to judge Claude’s SEO advice is to ask what kind of evidence the recommendation is based on.
I find it useful to think about Claude’s answers in three buckets.
Observed
Observed means Claude directly had access to the information it’s describing.
For example: “This page has an H1 that says ‘Pelvic Floor Physical Therapy in Seattle.'”
Claude saw the page and read the heading. That’s an observation.
Inferred
Inferred means Claude is making a reasonable judgment based on the information it can see.
For example: “‘Pelvic floor therapist Seattle’ may be a useful search phrase to investigate.”
That might be a smart idea. But unless Claude has actual search data, it’s still an idea to investigate.
Data-backed
Data-backed means Claude has been given an external source that verifies the claim.
For example: “The Google Search Console data you provided shows 640 impressions for this query over the past three months.”
Now Claude isn’t guessing about whether people are searching that phrase. It’s working from actual performance data that you provided.
This is the question I’d get in the habit of asking whenever Claude gives you an SEO recommendation:
Did Claude observe this, infer this, or is it backed by data?
That one question will save you from taking a lot of confident-sounding advice too literally.
So, can Claude do your SEO?
Yes, with caveats. Claude can do a significant amount of SEO work well, especially analysis, comparison, content review, prioritization, drafting, and implementation support.
Using Claude for SEO works best when it has a clearly defined job, the information required to do it, and the methodology, criteria, context, and boundaries to work within.
I don’t expect Claude to decide the strategy from scratch.
Good Claude SEO is about giving Claude a precise job and making sure it has the evidence needed to do it.
Claude can implement the strategy. You still need the strategy
Claude can help implement a strategy. What it’s much less equipped to do on its own is decide which SEO problems actually matter most, which opportunities are worth pursuing, what should happen first, which pages should target which searches, where your site is losing visibility, and what work is most likely to lead to more traffic and inquiries.
That’s the strategy piece.
My SEO Site Review & Strategy Package is where I figure that out for you.
I review your website, search performance, opportunities, and priorities, then give you a clear plan for what to fix, what to optimize, and what to focus on next.
From there, Claude can be a very capable implementation partner.
If you want to talk through what that could look like for your website, book a call with me.
