Auditing a site with Claude: the right method (and why AI alone gets it wrong)

You have a site, you have access to Claude, and the idea looks obvious: copy your site's address, paste it into the conversation, and ask for a full audit. In seconds you get a list of recommendations. The problem is that the list is almost always the wrong one.

Asking Claude to audit a site with no framing at all is the worst way to use an excellent tool. Left to itself, it concentrates on what it can see easily: image alt tags, meta descriptions, title lengths. Those are real elements, but they are rarely the ones that explain why a site does not generate clients. Worse, with no context it applies generic methods that match neither your goals, nor your market, nor the reality of your firm.

This guide explains how to audit a site properly, why artificial intelligence is a poor autonomous auditor but an excellent executor when it is well framed, and which questions you have to ask yourself before you even open a conversation with Claude.

Why does asking Claude for an audit with no context give bad results?

What happens when you hand a raw site to an AI?

When you give Claude a site's address and ask it to "audit this site", you are asking it to assess something whose purpose it does not know. A site is not good or bad in the abstract: it performs or it does not, against a goal. And the goal is exactly what the AI does not know. So it falls back on universal criteria that can be measured without context, that is, the least important criteria.

The result is predictable. Claude tells you that this image has no alt text, that that meta description is 180 characters instead of 155, that this page could have a catchier title. You come away with a list of thirty cosmetic micro-corrections, and you miss the only question that counts: can this site reach the goal you set it, in the market you are in?

Why can good methods not be invented without a frame?

Auditing a site properly assumes a method, and a method assumes a starting point. With no defined goal, the AI has no way of prioritising. It cannot know that for a firm aiming only at brand awareness, SEO structure matters little, whereas for a firm wanting to rank on competitive queries it is everything. The same site deserves two radically different audits depending on what is expected of it. That is exactly what the AI cannot guess on its own.

What is the first thing to do before any audit?

Define the site's goals

Before thinking about corrections, ask yourself a simple question: what is this site for? The possible answers are very different from one another, and each calls for a distinct audit.

Some sites exist only to exist: to be found when someone types the lawyer's name or the firm's name, to serve as an online business card, to reassure a prospect who already has the contact. That is a perfectly legitimate goal, and it is in fact the base. If you type your name and your firm's name and your site comes out first, that foundation is secured. For that kind of goal, most of the work lies somewhere other than in heavy SEO optimisation.

Other sites have an ambition to win ground: to rank on queries of the type "lawyer + specialism + city", to capture a demand prospects express without knowing your name. That is an entirely different goal, far more demanding, and it changes the nature of the audit completely.

Understanding what is inside the site

The second question matters just as much: what is actually in this site? How many pages, which pages, what content, what depth? A five-page site and a two-hundred-page site are not diagnosed the same way. Most of the time it is not the site itself that is the problem, it is the quantity and the quality of the content in it. A technically perfect site empty of useful content will never rank on competitive queries.

Why should you assess what is actually possible before auditing?

A site alone is not enough to rank on competitive queries

Here is the most frequent trap. A firm has a beautiful, well-built site, and is surprised not to come up on "employment lawyer Paris". The disappointment comes from a badly calibrated expectation. Ranking on a broad query, in a large city, against dozens of competitors who have been producing content for years, is a goal you do not win with a site "alone", meaning without a serious search strategy and without regular content production.

What is possible depends on your starting point. On your name and your firm, the win is close to automatic. On a competitive, geolocated trade query, you need a content strategy over several months, dedicated pages, authority built up. Before asking Claude to audit anything, you have to know which category your ambition falls into, or the audit will lead you to recommendations unrelated to what is actually reachable.

How do you frame an audit with Claude properly?

Share the sitemap so it can read the structure

If you want a useful audit, give Claude the means to understand the structure. Share the sitemap: that is what lets it see the overall architecture, the number of pages, how they are organised, rather than discovering the site page by page with no overall view. A sitemap properly handed over transforms the reading and allows a structural analysis instead of a succession of isolated observations.

Give your goals, or the audit is useless

This is the central point of the whole guide. With no explicit goals, the audit is worth nothing. Tell Claude clearly what you expect from the site: brand awareness, ranking on precise queries, generating enquiries, local presence. Give it your market, your area, your competitors if you know them. The more precise the frame, the more relevant the analysis becomes. That is the difference between "here is a generic list" and "here is what is actually blocking your goal".

Which optimisations have little impact, and which genuinely count?

The low-return optimisations

Some tasks take a lot of time for a marginal effect. Image alt text and meta descriptions are among them. They are important elements, and the point is not to ignore them: a well-written meta description improves the click-through rate from the search results, which has an indirect effect. But correcting them first is confusing the incidental with the essential. They are not what decides your ranking.

The genuinely decisive elements

The high-impact levers are of another kind. Meta title tags weigh heavily on ranking and on clicks. The heading structure (the H1 down to the subheadings) shapes how the engines understand your pages. The canonical tag avoids the duplicate content problems that can sabotage a whole site. And above all, the existence of a dedicated page per query: to rank on a subject you need a page that treats that subject in depth, not a mention buried in a generalist page. That is where performance is decided, and that is what an audit has to examine first.

Navigation, a major factor the AI assesses badly

Navigation matters a great deal: it distributes authority between your pages, it guides the visitor and it structures how Google understands the site. It is also one of the points Claude will assess with the most difficulty, because the quality of a navigation is judged in use, on the real experience of a visitor looking for something, not on a static reading of the code. It is exactly the kind of element where human judgement stays essential.

Why have the quality and the clarity of the content become decisive?

Content production is intensifying, and competition with it. The barriers to entry are now completely broken: anyone can produce articles in volume at low cost. The consequence is paradoxical: the more content there is, the more quality becomes the only point of differentiation. Publishing is no longer enough, you have to publish better.

Concretely, that means concentrating as much on the quality as on the clarity of the article and its real usefulness to the reader. An article that answers precisely a question a prospect is asking, that teaches them something, that helps them decide, will beat ten generic articles produced on a production line. That is the ground on which search is now won, and it is also what a purely technical audit will never measure.

What role does the site-building tool play in performance?

The limits of certain site builders

Not all site-building tools are equal on performance. Some platforms, often the ones that generate sites through AI such as Wix or Squarespace, impose technical limits you risk running into: throttled performance, restricted access to the code, constraints on the structure. Those are points an audit has to identify, because they cap what you will be able to achieve, whatever the quality of your content.

Building a site with AI can work perfectly well

That does not condemn AI-assisted site building, which can succeed perfectly well in search. The condition is to treat SEO as a central element of your project management, from the start, and not as an afterthought. That means folding search into the design of the site itself, creating dedicated skills to frame the AI's work, and checking through Google Search Console that your pages index quickly. Plenty of people hit a wall on exactly that point: a site that does not index is invisible, whatever its apparent quality. It is not at all fatal, but it is a real risk to watch from the first days.

How long should you wait before judging the results?

One last thing, and it conditions all the others. SEO is slow. After a month you will already have plenty of information about whether the trajectory is going the right way: pages indexed, first impressions, first keywords appearing. But the figures will be very low, and that is normal. Judging a site on its results after a few weeks is the surest way to take a bad decision.

Six months of latitude is genuinely necessary to assess a search strategy. That is the period after which content settles, gains authority and starts producing measurable results. An audit that does not account for that horizon will push you to correct things that simply had not had time to work yet.

How does OURAMA approach auditing a firm's site?

At OURAMA, an audit never starts with the list of tags to correct. It starts with the firm's goals and with an honest reading of what is possible: can this site, in this market, with this content, reach what the firm expects of it? Only once that question is settled does the technical analysis take on meaning, because it is then ranked against a real purpose.

We use artificial intelligence daily, but as a framed executor, never as an autonomous auditor. The judgement on navigation, on the relevance of the positioning, on the real quality of the content, on a reasonable time horizon, stays human work, informed by a knowledge of the regulated professions market we have built across more than 400 firms.

If you want to know where your site actually stands against your goals, and what can reasonably be expected of it, we run complete audits with no jargon. Get in touch.

‍

Questions people ask most about auditing a site with Claude

Can you genuinely audit a site with Claude?

Yes, provided you frame it. Claude becomes useful once you give it the sitemap, the site's precise goals and the market context. Without that frame it concentrates on minor elements and applies generic methods. With it, it becomes an excellent executor able to analyse a structure, spot inconsistencies and propose ranked corrections. The rule is simple: AI executes well what you ask it precisely, and judges badly what you leave it to interpret alone.

Which optimisations count most for a law firm site's search ranking?

The high-impact levers are the meta title tags, the heading structure, the canonical tag to avoid duplicate content, and the existence of a dedicated page per targeted query. The quality and the clarity of the content have become the main point of differentiation. Conversely, alt text and meta descriptions, useful as they are, have a secondary impact and should not be handled first.

How long before you see the results of SEO work?

Count roughly a month for the first signals of trajectory (indexing, first impressions, keywords appearing), but with figures still very low. A serious assessment needs six months of latitude, the time for content to gain authority and produce measurable results. Judging a site over a few weeks almost always leads to bad decisions.

Can a site built with an AI tool such as Wix or Squarespace rank well?

Yes, but with vigilance. Those platforms sometimes impose technical limits (throttled performance, restricted access to the code) you can find yourself stuck with. The key is to fold SEO into the project from the design stage, to create dedicated working frames and to check quickly through Google Search Console that the pages index. The main pitfall is indexing: a site that does not index stays invisible. It is not fatal, but it is a risk to watch from the start.

Do you need a heavily optimised site to appear on your own name?

No. Ranking on your name and your firm's name is the base, and a properly built site gets there almost automatically. It is only for competitive trade queries (of the type "lawyer + specialism + city") that a serious search strategy and regular content production become essential. So it all depends on the goal you set your site.