AI website audit vs hiring a consultant
Two companies with opposite incentives both measured what automation catches. They got 13% and 57%. Both are right, and the reason they differ tells you which to buy first.
An automated audit and a hired consultant do not find the same things, and the gap between them has actually been measured. Not in conversion work, where nobody publishes, but in accessibility testing, where two companies with opposite incentives both did the maths.
If the choice on the table is an ongoing retainer rather than a one-off consultant, the deciding number is how many conversions your site produces, not how much traffic. That threshold is in is a CRO agency worth it.
Their answers were 13% and 57%. Both are correct. The reason they differ tells you exactly which one to buy first.
What can an AI website audit actually find?
Automated audits reliably find anything countable or measurable: missing labels, unlabelled buttons, contrast failures, slow-loading assets, absent trust signals, a form with fourteen required fields. They cannot judge whether your headline is persuasive or your offer is worth the price.
That split is not a limitation of any particular tool. It is the difference between measuring a thing and having an opinion about it. A scanner can tell you an image has no alt text. It cannot tell you the alt text that is there says "photo".
Where the two published numbers disagree
Accessible.org rates 13% of WCAG 2.2 AA criteria as reliably machine-detectable, 45% partially, and 42% not at all. Deque, analysing 294,958 real issues, found automation caught 57.38%. Neither is wrong. They count different things.
One counts criteria, the kinds of problem that exist. The other counts issues, the individual instances on real pages. Automation looks weak by the first measure and strong by the second, because the problems it can catch happen to be the ones that occur constantly.
- 13% of criteria reliably machine-detectable, 7 of 55
- 42% of criteria not machine-detectable at all
- 57% of actual issues still caught by automation
Both numbers come from companies with something to sell, in opposite directions. Accessible.org sells manual audits and publishes the criterion-by-criterion table so you can check it. Deque sells the automated tool and says so openly, then publishes the full issue counts. Read them against each other and the truth sits in the middle.
The number hiding inside the 57%
Deque's own table shows colour contrast made up 30% of every issue found, and automation caught 83% of it. Strip that one check out and the average collapses. Keyboard access was caught 2.49% of the time. Focus order, zero.
So the honest reading is not "automation covers half the work". It is that automation covers one kind of work almost completely and another kind not at all, and the headline average depends entirely on which kind your site has more of.
Wondering which kind yours has? Run a free audit and see what a machine can and cannot tell you.
The question was never tool or human. It is which half of the problem you have.
What only a person will catch
A human catches anything requiring a judgement about meaning: whether your value proposition is understood in five seconds, whether your pricing page invites a comparison you lose, whether the offer itself is weak. No tool has an opinion on whether your product is worth buying.
Concretely, the things a consultant finds that an audit will not:
- A weak offer. Every element on the page can be correct and the thing being sold still not worth the money.
- The wrong audience. A page that converts badly because the traffic was never going to buy is a targeting problem wearing a conversion costume.
- Positioning. Whether you are the obvious choice for someone, or a slightly worse version of a competitor.
- Sequence. Whether the order you reveal price, proof and promise matches how this particular buyer decides.
Which should you buy first?
Buy the automated audit first, almost always. It costs a fraction of a consultant, it takes minutes, and it clears the countable problems off the table so a consultant's hours go to the judgement calls rather than to finding your unlabelled buttons.
If the choice is between tools rather than between a tool and a person, Revslip alternatives says which rival to buy instead and when.
Published 2026 rates put experienced CRO consultants at $100 to $300 an hour, specialists with documented results at $200 to $300, monthly retainers between $2,500 and $10,000 for small and mid-sized businesses, and a full funnel audit at $5,000 to $15,000, according to one agency rate guide. Treat that source the way we asked you to treat ours, it is an agency quoting the market it sells into.
At $200 an hour, a consultant spending their first day cataloguing missing labels and slow images costs you $1,600 to produce a list a scanner generates in four minutes. That is the whole argument for sequencing. Machines first, because they are cheap and exhaustive on the countable half. People second, because their expensive hours should go somewhere a machine cannot reach.
Where we would push back on ourselves Revslip sells the automated half, so treat the recommendation above with the suspicion it deserves. The accessibility numbers are an analogy, not a measurement of conversion tools. Nobody has run the equivalent study on CRO audits, including us, and we are not going to pretend our own 134-site sample settles it.
When the consultant goes first instead
Skip straight to a person when the problem is upstream of the page. If you are not sure who the product is for, if you have never had a customer explain why they bought, or if you are choosing between two business models, an audit will hand you a tidy list of fixes to a page that should not exist yet.
Same if you have already fixed the mechanical problems. A second scan of a clean site returns very little, and at that point the remaining money is in judgement, which is what you are paying a human for.
The tell is whether your open questions are about the page or about the business. "Why does this form lose people" is a scanner question. "Why does anyone choose us" is not.
Common questions
Is an AI audit the same as pasting my URL into ChatGPT?
No. A general chat model cannot load your page, measure anything, or see what a visitor sees, so it produces plausible advice about a site it never visited. The difference is worth understanding before you trust either.
How much does a proper CRO audit cost?
Published bands run from free automated scans to five figures for agency work. The real price bands and what changes between them covers where each tier stops being worth it.
Can I just use a free audit tool?
Often yes, as a first pass. What varies is what each one actually checks, which is rarely stated. Here is what free tools do and do not look at.
How many things can an automated conversion audit check?
Revslip runs about 200 signals across copy, offer, trust, UX, technical and tracking, and prices each finding. The conversion leak index lists what those categories contain. Our audit reports them ranked by estimated monthly cost.
If you would rather start with the countable half, paste your URL and see what comes back.