An SEO audit is the analysis that lists the reasons a site does not rank as well as it could, puts them in order of impact and assigns each one to whoever has to fix it. An automated crawler can do the first two. It cannot do the third, and the third is what decides whether the audit will be any use or end up as a PDF in a shared folder.
The difference between an audit and an export of errors is all in the order. Run Screaming Frog on a three-thousand-page site and you get a few thousand warnings: missing alt text, redirect chains, duplicate meta descriptions. Almost none of those lines, on its own, moves a single position.
Below you will find three things we did not find on the first page of Google UK: data on which technical problems really exist across the web, a check we ran on 26 September 2026 on the nine pages that rank in the UK for "seo audit", and the criterion we use to decide what goes on the list of fixes and what stays off it.
What an audit finds when you run it on the whole web
The HTTP Archive project crawls millions of pages every year and publishes the results in the Web Almanac. The SEO chapter of the 2025 edition is the quickest way to see which errors are common and which are rare, instead of trusting how sensitive your crawler happens to be.
| Element | Mobile pages that have it |
|---|---|
| Title tag | 98.54% |
| robots.txt returning 200 | 84.9% |
| Meta description | 67.2% |
| Non-empty H1 | 66% |
| Canonical tag in the raw HTML | 64.27% |
| Alt attribute on images (median) | 60% |
| Core Web Vitals in the good range | 48% |
| llms.txt | 2.10% |
Everyone has a title tag: as an audit item it is worth nothing, unless you turn out to be in the 1.46% without one. The meta description is missing on a third of pages and is almost always the first warning a tool puts at the top. It is not a ranking factor and Google often rewrites it anyway: it is a click problem, and clicks are checked in Search Console, not in the crawler report. If you want the detail on how to write a meta description that holds up, we covered it separately.
The row that deserves attention is another one. 64.27% of pages have a canonical in the raw HTML, and Core Web Vitals are in the good range for only 48% of mobile pages. These are two areas where a real problem is statistically likely, not a textbook hypothesis.
We ran the checklist on the tools that sell it
On 26 September 2026 we took the nine organic results that rank in the UK for "seo audit" (all nine are audit tool pages: SEOptimer, Semrush, Ubersuggest, SEO Site Checkup, The HOTH, Ahrefs, Seobility, AIOSEO, SEOmator) and downloaded them with a server-side HTTP request, using a user agent that does not pretend to be a browser. Then we checked the same items those tools check: response code, title, meta description, number of H1s, canonical, structured data, robots.txt, AI crawlers, llms.txt.
| Check | Result |
|---|---|
| Page reachable by a non-browser client | 9 of 9 |
| A single H1 | 8 of 9 |
| Canonical present | 9 of 9 |
| Meta description present | 9 of 9 |
| Title within 60 characters | 6 of 9 |
| FAQPage structured data | 5 of 9 |
| robots.txt reachable | 9 of 9 |
| robots.txt with a Sitemap directive | 8 of 9 |
| At least one AI crawler named in robots.txt | 4 of 9 |
| llms.txt present | 5 of 9 |
The basics are all in place: every page answered, all nine have a canonical and a meta description, eight have a single H1. It is worth comparing with the same check we ran on the Italian first page for the same query on 7 September 2026, which was made of agency guides: there, three pages out of ten returned 403 to a request that did not identify itself as a browser. Those sites almost certainly let Googlebot through, since it is verified by IP range and not by user agent string. The point is that everything else is kept out, and "everything else" now includes the crawlers that feed generative answers.
On the AI side the tool vendors are well ahead of the web: five of the nine publish an llms.txt and four name at least one AI crawler in robots.txt, against 2.10% of pages with llms.txt in the Web Almanac. The companies that sell the checklist apply it. The sites you will actually audit look much more like the Web Almanac than like this table.
The work is not finding the errors, it is ordering them
Stephanie Wallace set out in Search Engine Land, in May 2026, the most concise criterion we have read on the subject (How to prioritize technical SEO fixes by business impact): before putting a warning on the list, answer three questions. Does it affect crawling or indexing? Does it hit pages that are worth something? Is there evidence that it is suppressing traffic or rankings? If the answer is no three times, that line is not a priority, however red the tool marks it.
This is the grid we use to set the order, and it should be read knowing that the order changes with the type of site.
| Fix | Moves traffic | Effort | When it comes first |
|---|---|---|---|
| Important pages set to noindex or blocked by robots.txt | Yes, within days | Low | Always, before anything else |
| Duplicate versions of the site not redirected (www, http, staging) | Yes | Low | Always |
| 404s on URLs that receive external links | Yes | Low | If the domain has backlinks |
| Missing title and H1 on the pages that generate leads | Yes | Low | Always |
| Wrong intent on pages stuck in positions 8-20 | Yes, more than anything else | Medium | Almost always |
| Two pages competing for the same query | Yes | Medium | If cannibalisation is confirmed in Search Console |
| Core Web Vitals outside thresholds | Little, if the site is already usable | High | After the rest, or if the site is visibly slow |
| Internal redirect chains | Little | Medium | On sites above tens of thousands of URLs |
| Alt text on decorative images | No | Low | Never as a priority, it remains an accessibility matter |
| Crawl budget optimisation | No, below 10,000 pages | High | See the section below |
The first four rows are the only part of the audit that must be closed within the week. Everything else can wait, and should be written in an order that someone can actually carry out. For how to sequence the on-page work afterwards, the on-page SEO guide goes through it item by item.
Crawl budget is almost certainly not your problem
It is the item that appears in almost every audit template and in nine cases out of ten it is wasted time. Google has stated in writing, in its guide to managing crawl budget for large sites, which sites need to worry about it: "Large sites (1 million+ unique pages) with content that changes moderately often (once a week)" and "Medium or larger sites (10,000+ unique pages) with very rapidly changing content (daily)". Below those thresholds, if new pages are crawled on the day you publish them, crawl budget optimisation does not concern you.
It helps to keep the scale of the market in mind. According to the Business population estimates 2025 from the Department for Business and Trade, the UK has 5.7 million private sector businesses, and 75% of them have no employees other than the owners. The vast majority of the websites behind those businesses sit two or three orders of magnitude below Google's threshold. Spending half an audit on crawl budget for a four-hundred-URL site means the outline was copied from an article written for enterprise ecommerce.
Pages in positions 8-20 are the chapter everyone leaves out
No crawler does this part, and it is the part of the audit that pays back first. The Performance report in Google Search Console, filtered on the queries where the site appears between position 8 and 20, gives you the list of pages that Google already considers relevant and that sit just outside the first screen.
On those pages the problem is hardly ever technical. In most of the cases we see it is a mismatch of search intent: the page answers a nearby question, not the one that was typed. An audit that does not open this report delivers a list of technical problems that is correct and useless.
Three checks we always run on this group of URLs, in this order: the main query appears in the title and in the first paragraph, the page covers the subtopics covered by the top three results, and there is no second page on the site competing for the same query. The third case is more common than people think and is the only one of the three that requires a decision: merge, differentiate or add a canonical. We have gone into how duplicate content works elsewhere.
The audit now looks beyond Google too
In the Web Almanac 2025 GPTBot appears in 4.5% of desktop robots.txt files and llms.txt exists on 2.13% of sites. The numbers are low, and our check shows how far apart the two groups are: of the nine UK pages ranking for "seo audit", four name an AI crawler and five have an llms.txt (on the Italian first page for the same query, one and two out of ten). The businesses that sell audits have moved; most of the sites they audit have not.
We have added four items to our audits since generative answers started to weigh on traffic, and none of them needs new software:
- Check that the firewall or CDN is not returning 403 to clients that do not identify as browsers, because that is how you lose citations in AI answers without noticing.
- Check which AI crawlers are allowed in robots.txt, and that this is a decision someone made rather than a plugin default.
- Look at whether the content that matters is in the HTML served or only appears after JavaScript runs.
- Open the AI features reports in Search Console, which since June 2026 separate impressions coming from AI Overviews and AI Mode.
Structured data stays on the list, but with a different hierarchy from five years ago: it helps you be read without ambiguity more than it wins a rich result. For the full picture on optimising for generative engines, see the guide to generative engine optimisation.
A case where what moved the rankings was not technical
Macropix makes custom LED screens in Milan. When the project started, in 2020, the site was in position 88 for "monitor pubblicitario" (advertising monitor) and in position 73 for "schermo pubblicitario" (advertising screen): pages nine and eight of Google, in other words zero traffic. In 2025 both queries are in position 2, and on the LED wall cluster the domain has a 25.55% share of voice, ahead of amazon.it at 18.08%.
The part that matters for audits is the one usually left out when these numbers are told: none of those jumps came from a technical fix. They came from moving the technical content out of catalogue PDFs and giving each product line its own page. The full table of the nine keywords, with volumes and positions for 2020 against 2025, is in the case study on SEO for manufacturing. Other projects with the numbers in full are on the projects page.
How to deliver an audit that someone actually applies
A technical audit we delivered to an international consultancy firm in 2021 was divided into six chapters: introduction and objectives, crawling and indexing, performance, site structure, content, cross-cutting optimisations. The structure still holds today. What has changed is what we put in each line.
Four columns for each fix: what problem it solves, who carries it out (development, editorial, marketing), how many hours it costs, what effect we expect and on which URLs. Without the second column the audit never starts, because nobody knows whether that line is theirs. Without the fourth nothing can be verified three months later, and an audit that cannot be verified is a well-formatted opinion.
The cadence we recommend: a full audit when a project starts, then a lighter check every six months, and an extra one after a migration, a redesign or an unexplained drop in traffic. Redoing the full analysis every quarter only benefits whoever sells it.
What an SEO audit costs in the UK
Prices published by UK agencies give an idea of the range. Whitehat SEO states £500 to £7,500; Rubik Digital £1,000 to £5,000; Red Eagle Tech £500 to £2,500 for sites under 100 pages and up to £15,000 and more for sites above 500. These are stated figures, not a market survey, and should be read that way.
The variable that moves the price is not the number of pages, it is how much of the diagnosis can be automated. A crawl and a generated report take a few hours. Reading Search Console query by query, checking positions 8-20 and deciding what to merge and what to rewrite is manual work that does not go below a certain threshold. If an audit quote costs as much as two hours of work, what you receive is the tool's export. On the relationship between price and actual work we wrote about how much SEO costs, and realistic timings are in how long SEO takes to show results.
If you would rather have the analysis done from outside, the starting point is our SEO services page: the audit is the first thing we deliver on every project, before any fix.
Frequently asked questions
A full one when a project starts, then a lighter check every six months. An extra one is needed after a migration, a redesign, a change of CMS or a drop in traffic that seasonality does not explain. Redoing the full analysis every quarter produces no new information on a site that has not changed.
The diagnostic part, yes. Google Search Console is free and covers indexing, queries, positions and crawl problems; Screaming Frog crawls up to 500 URLs without a licence. The hard part is not collecting the data, it is deciding which of the problems found are worth fixing first: that is where experience makes the difference in cost.
Three cover 90% of the work: Google Search Console for real search data, a crawler such as Screaming Frog or Sitebulb for crawling, PageSpeed Insights for performance. Suites such as Semrush, Ahrefs or Sistrix are for backlinks, ranking history and competitor comparison, not for technical diagnosis.
No, unless the site has a real block, such as important pages set to noindex or a duplicate version that is not redirected. Fixing that block produces a quick jump. Outside those cases growth comes from content and from pages stuck in positions 8-20, not from reducing the number of warnings in the crawler report.
Crawling and data collection take hours. The manual analysis on a site of a few hundred pages usually takes between two and four working days, and it grows with the number of different templates to check, not with the number of URLs: a hundred product pages built on the same template are checked once.
One thing the industry rarely says: the most useful part of an audit is the list of what we decided not to do. The sixty discarded lines, each with its reason. It helps whoever reads the audit six months later and finds the same warning in the crawler report, and above all it helps when the next consultant arrives, opens the same tool and proposes to fix everything. That list is the only thing that separates a decision from an oversight, and we have never seen it attached to an audit done by someone else.
