The technical SEO audit checklist: every check, in order.

1. How to run this audit.

Run the groups in the order they appear, because that is the order in which the pipeline can break. A page has to be crawlable before it can be indexed, indexed before rendering matters, and rendered before anybody argues about layout shift. Auditing performance on a page that is excluded by robots.txt is a way to spend an afternoon producing a report that is entirely correct and completely useless.

Every check here is paired with a signal: the observable thing that tells you this specific item failed. That pairing is the difference between a checklist you can run on a site you built and one you can run on a site somebody handed you an hour ago. Work from the symptom back to the check rather than reading all 38 in sequence, and a site with a real problem usually resolves in under an hour.

One boundary, because these two pages get conflated constantly. This is the line-by-line technical audit. The whole-program version, covering setup, keyword selection, on-page structure, links, and measurement with the technical layer compressed into a seven-check floor, is our complete SEO checklist. Use that one to run a content operation. Use this one when something that used to work has stopped, or a section will not index, and you need to find the break. If you want the machine-generated version of this pass over your own domain first, the site audit report crawls the site and returns the findings grouped roughly the way this page is.

Last framing note. Almost nothing on this list is a ranking lever. These are disqualifiers: a correct canonical tag will not move you a position, and a missing one can cost you every position on that URL. That distinction sets the expectation for what a clean audit means. Ahrefs studied roughly a billion pages and found 96.55% of them get zero organic search traffic from Google, and the technical layer is the reason for a small minority of that number. If the audit comes back clean and the pages still earn nothing, the audit did its job and the problem is upstream in what the pages are for.

2. Crawl access and discovery.

Eight checks on whether a crawler can reach the page at all. This group holds the cheapest bug in the discipline, a single disallow line left over from a staging build, and it is worth reading the file rather than trusting a summary of it. Google Search Central is explicit that robots.txt controls crawling, not indexing, which is the distinction that trips people up: a blocked URL can still appear in results, stripped of everything that would have made it worth clicking.

1. 01

   robots.txt blocks nothing you need crawled.

   Read every line rather than scanning it. A staging disallow that survived launch is the cheapest possible bug and the most expensive one to leave in place. Test the specific URL, not the directory.

   SIGNAL Pages excluded by robots.txt in Search Console, on URLs you want ranked.

2. 02

   Nothing important is blocked that the page needs to render.

   Blocking the CSS and JS bundles a page needs is a slower version of blocking the page. The crawler fetches a layout it cannot build and grades what it gets.

   SIGNAL Rendered screenshot in the URL inspector looks nothing like the live page.

3. 03

   The XML sitemap lists only indexable, canonical, 200-status URLs.

   Redirected URLs, noindex pages, and 404s in a sitemap are a contradiction: you are requesting a crawl of a page you already told search engines to skip. Split large sitemaps by section so coverage errors point somewhere.

   SIGNAL Sitemap submitted count far above the indexed count, with errors in the coverage report.

4. 04

   Every indexable page is reachable from a crawlable HTML link.

   Reachable means an anchor tag with an href. A URL discoverable only through a sitemap, a search box, or a click handler is a page the crawler has to be told about rather than one it finds.

   SIGNAL Orphans in a crawl of your own site: URLs present in the sitemap, absent from the link graph.

5. 05

   Important pages sit within three clicks of the homepage.

   Depth is a proxy for how much the site itself thinks a page matters. Money pages buried six levels down behind pagination get crawled rarely and valued accordingly.

   SIGNAL Crawl depth report shows commercial URLs at depth five or more.

6. 06

   The crawl is not being spent on parameter and filter URLs.

   Faceted navigation generates a near-infinite URL space out of a finite catalogue. Decide which parameter combinations are allowed to exist as crawlable URLs, then enforce it.

   SIGNAL Server logs dominated by requests for filtered and sorted variants of the same listing.

7. 07

   Crawl budget is only treated as a constraint at the size where it is one.

   On a site of a few thousand pages, crawl budget is almost never the reason something is not indexed. Chasing it on a small site is the most common way an audit spends a week and moves nothing.

   SIGNAL Nothing. This one is a check against inventing a problem you do not have.

8. 08

   The pages you care about are actually indexed, confirmed one by one.

   Live, linked, in the sitemap, and crawled are four separate states, and none of them is indexed. Only the last one can rank, so confirm it directly rather than inferring it.

   SIGNAL Crawled currently not indexed or Discovered currently not indexed in the coverage report.

Check seven is a check against inventing work, and it deserves the space. Crawl budget is a real constraint on large sites and almost never the explanation on small ones. Google's own guidance on managing crawl budget for large sites scopes it to sites with over a million pages that change weekly, or over ten thousand that change daily. Below that, a page that is not indexed was not skipped for budget reasons, and the answer is somewhere in the next group.

Discovery is the half of this group that gets skipped because it does not feel technical. A page nothing links to is found late, valued little, and usually forgotten by whoever wrote it inside a week. Fixing that is a mapping problem rather than an infrastructure one, which is why our internal link mapper exists: it shows you what points at what, and the orphans fall out of the graph immediately.

3. Indexation and duplication.

Six checks, and this is where most audits actually end. Pull the list of URLs that carry commercial intent, confirm each one is in the index, and about half the time the coverage report explains the entire problem on its own. The other half of this group is duplication, which is the same question asked from the other direction: how many near-identical URLs are competing for one position.

1. 09

   No stray noindex left on a template.

   These are added during a build and removed from the page nobody re-checked. Inspect the rendered head of a real production URL, not the source of the template, and check the HTTP header too.

   SIGNAL Excluded by noindex tag on URLs you expect to rank.

2. 10

   The index is not full of pages you never meant to publish.

   Tag archives, author pages, internal search results, print views, and thin paginated pages. Each one is a thin page competing with your real ones for the same attention.

   SIGNAL Indexed page count several times larger than the number of pages you would defend.

3. 11

   Every URL has a self-referencing canonical.

   It costs nothing, survives tracking parameters and syndication, and removes the ambiguity that turns into a duplication problem the first time somebody shares a link with a campaign tag on it.

   SIGNAL Duplicate without user-selected canonical, or Google chose a different canonical.

4. 12

   No canonical points at a redirected, noindexed, or 404 URL.

   A canonical aimed at a dead target is a contradictory instruction, and the resolution is not in your favour. This breaks silently during migrations and nobody notices for months.

   SIGNAL Canonical target returning anything other than a 200 in a site crawl.

5. 13

   One canonical hostname, every other variant 301ed to it.

   www or bare, http or https, trailing slash or not. Four live variants of one page is the classic way a site splits its own signal four ways and wins with none of them.

   SIGNAL More than one hostname variant returning a 200 instead of a redirect.

6. 14

   Titles and meta descriptions are unique across the archive.

   Duplicates cluster where pages are templated: paginated archives, location pages, product variants. They are a symptom of pages that are too similar to each other, not just a tagging problem.

   SIGNAL Duplicate title tags grouped in a site crawl, almost always in one template.

The canonical checks come with a caveat worth stating plainly. Google's documentation on consolidating duplicate URLs describes the canonical link element as a signal rather than a directive: it can be overruled when other signals disagree. So do not treat a canonical as a fix for two pages that genuinely target the same query. Merge them, redirect one at the other, and you have solved the actual problem instead of labelling it.

Index bloat is the check that most often surprises people. A site with 300 real pages and 4,000 indexed URLs is not benefiting from the other 3,700, and the archive templates producing them were nobody's decision. The structural fix is to decide which templated URL patterns deserve to exist as indexable pages, which is the same discipline behind building page templates at scale: a template earns its index slot by answering a distinct question, or it should not be generating URLs.

4. Status codes and redirects.

Six checks on what the server says when asked. This group is almost entirely migration debris: chains that grew a hop per replatform, 302s from a framework default, and mass redirects to the homepage from a cleanup somebody did in a hurry. Google's guidance on redirects and Google Search is clear that a server-side 301 is the strongest available signal for a permanent move, and that a redirect to an irrelevant page is treated as a soft 404.

1. 15

   Permanent moves use a 301, not a 302.

   A temporary redirect tells search engines to keep the old URL. Leave it in place for a year after a permanent migration and you have spent a year telling them the move has not happened.

   SIGNAL Long-lived 302s in a crawl, usually from a platform default nobody changed.

2. 16

   No redirect chains longer than one hop.

   Chains accumulate over migrations: http to https to www to the new slug. Each hop is latency and each one is an opportunity for a link in the chain to break. Repoint the first URL at the final destination.

   SIGNAL Redirect chain and redirect loop reports in a site crawl.

3. 17

   Internal links point at final URLs, not at redirects.

   After a slug change the redirect handles outside links. Your own links should be edited to the new URL, because every internal redirect is a self-inflicted hop on a path you control completely.

   SIGNAL Internal links to 3xx URLs, usually in a nav or footer template.

4. 18

   Removed pages return 410 or 404, not a soft 200.

   A page that says nothing found in the body while returning a 200 status is a soft 404: the crawler keeps requesting it and the index keeps a page with no content. Say what you mean in the status code.

   SIGNAL Soft 404 in the coverage report, or thin pages indexed at removed URLs.

5. 19

   Retired URLs redirect to the closest equivalent, not all to the homepage.

   A mass redirect of dozens of specific pages to the homepage is treated as a soft 404, because it is one. Map each URL to the page that answers the same question, and 410 the ones with no equivalent.

   SIGNAL A redirect map where one destination appears hundreds of times.

6. 20

   No 5xx errors under crawl load.

   Intermittent server errors during a crawl are the fastest way to have your crawl rate reduced. Check the error rate on the crawler user agent specifically, not just the site average.

   SIGNAL Server error 5xx in the coverage report, clustered by time rather than by URL.

The internal-links check is the one people argue about, so here is the reasoning. An external link pointing at an old URL is out of your control and the redirect is doing exactly its job. An internal link pointing at an old URL is a hop you chose, on a path you own completely, repeated on every page in the template. It costs nothing to fix and it removes a whole class of future breakage when somebody eventually prunes the redirect table.

5. Rendering and mobile parity.

Five checks on whether the page a crawler sees resembles the page a person sees. The mechanism is worth understanding rather than memorising: Google processes JavaScript in a deferred rendering pass, separate from the initial crawl. Content that only exists after that pass is still eligible, it just arrives later and depends on a second step completing. Content in the server HTML skips the queue entirely.

1. 21

   The main content exists in the HTML the server returns.

   View source, pick a sentence from the middle of the page, and search for it. If it is not there, indexation depends on a rendering pass that happens later and is not guaranteed to happen when you need it.

   SIGNAL Rendered HTML and raw HTML differ on the body copy, not just on widgets.

2. 22

   Internal links are anchor tags with real hrefs.

   A div with a click handler navigates a human and is invisible to a crawler. Routing that works entirely through JavaScript is the most common reason a well-built site has an orphan problem.

   SIGNAL A crawl that finds almost no internal links on a site clearly full of them.

3. 23

   The mobile page carries the same content as desktop.

   Indexing is mobile-first, so the mobile page is the page. Copy, headings, structured data, and links dropped from the small-screen layout are dropped from what gets indexed.

   SIGNAL Word count, heading count, or link count materially lower on the mobile render.

4. 24

   Nothing important is hidden behind an interaction the crawler will not perform.

   Content inside a tab or accordion is fine because it is in the DOM. Content fetched only after a click, a scroll, or a consent choice is not, because that fetch never happens.

   SIGNAL Sections missing from the rendered screenshot that are present for a real visitor.

5. 25

   Pagination and infinite scroll expose real, linkable URLs.

   If page two only exists as a scroll state, everything past the first screen of a listing is unreachable. Give every page of a sequence its own URL and link them in the markup.

   SIGNAL Deep listing items indexed nowhere while page one ranks fine.

Mobile parity is the check that catches the most surprising failures, because it is nobody's job. Google's mobile-first indexing documentation is unambiguous that the mobile version is the version used for indexing and ranking. A responsive layout that hides a sidebar of internal links below a breakpoint has removed those links from the index, and the desktop page everyone reviews still looks perfect.

If you publish through a CMS rather than a hand-built front end, most of this group is a theme-level question you answer once. That is the honest argument for shipping content through a native integration like our WordPress publishing setup rather than pasting markup into a page builder: the rendering behaviour of one reviewed template is a known quantity, and the rendering behaviour of 40 hand-built pages is 40 separate questions.

6. Core Web Vitals and page weight.

Six checks, and the only group on this page with published numeric thresholds, so use the real numbers rather than a general sense that the site feels quick: Core Web Vitals sets largest contentful paint at 2.5 seconds, interaction to next paint at 200 milliseconds, and cumulative layout shift at 0.1, measured at the 75th percentile of real visits. The percentile is the part people drop. It means a quarter of your visitors are allowed to have a worse experience than the threshold, and you are still passing.

1. 26

   Largest contentful paint under 2.5 seconds on field data.

   Field data means real visits at the 75th percentile, not a lab run on your laptop. A lab score of 98 on a wired connection tells you almost nothing about a phone on mobile data.

   SIGNAL The LCP group in the Core Web Vitals report sitting in needs improvement or poor.

2. 27

   The LCP element is the one you think it is.

   It is usually the hero image or the headline block. Teams optimise a carousel for weeks and find the actual LCP element was a web font blocking the H1 the entire time.

   SIGNAL Performance trace naming an element nobody expected as the LCP candidate.

3. 28

   Interaction to next paint under 200 milliseconds.

   This is the responsiveness metric, and on content sites the cause is nearly always third-party script on the main thread rather than your own application code.

   SIGNAL INP failing on mobile while passing on desktop, the standard third-party signature.

4. 29

   Cumulative layout shift under 0.1.

   Reserve height for images, ads, embeds, and anything injected after load. Most of the score comes from two or three elements, so fixing this is a short job once you know which ones.

   SIGNAL Content jumping when a banner, a font, or an embed arrives late.

5. 30

   Images are sized, compressed, and served in a modern format.

   An uncompressed hero is usually the single largest object on the page and the easiest win in the group. Serve real lossy WebP or AVIF at the dimensions the layout actually uses.

   SIGNAL One image several times heavier than the entire HTML document.

6. 31

   Third-party scripts are counted and justified.

   List every tag on the page and name what each one is for. Marketing tooling accumulates, nobody removes anything, and the performance work you do is quietly eaten by scripts nobody owns.

   SIGNAL More requests to other domains than to your own.

Keep this group in proportion. Performance is a tiebreaker between pages that are otherwise comparable, not a lever that lifts a weak page. The reason it absorbs so much audit attention is that it is the easiest layer to measure and produces the most satisfying numbers, which is a fact about dashboards rather than about search. If the three thresholds pass on field data, write it down and move to the next group rather than chasing a lab score of 100.

7. Structure, links, and markup.

Seven checks on how the page describes itself. These are the cheapest items on the list to fix and the ones most likely to be wrong at the template level, which means one fix propagates across every page using it. Run them on one URL per template rather than on a sample of pages, because the failures here are almost never page-specific.

1. 32

   One H1 per page, carrying the page topic.

   Templates that render a logo or a site name as an H1 produce two. Check the rendered page rather than the editor, because this is a theme-level bug, not a per-page one.

   SIGNAL Zero or multiple H1 elements in a crawl, consistently across one template.

2. 33

   Heading levels descend in order without skipping.

   Headings are the outline of the document. When they are picked for font size instead of hierarchy, the outline stops describing the page and both readers and parsers lose the structure.

   SIGNAL H4 elements appearing before any H2 on the same page.

3. 34

   Structured data validates with no errors on the type you claim.

   Ship the type the page actually is. Invalid markup is ignored at best, and markup that describes a page you do not have is the kind of thing that gets a site an entire class of features removed.

   SIGNAL Errors in the rich results test, or enhancements disappearing from Search Console.

4. 35

   Every page has a unique, descriptive title that leads with the topic.

   Around 60 characters before truncation. Impressions with no clicks is almost always a title problem rather than a content problem, which makes this the highest-leverage item in the group.

   SIGNAL Pages with impressions, a position inside the top twenty, and a click rate near zero.

5. 36

   Internal links are deliberate, varied in anchor, and point both ways.

   Most sites link in one direction only, so the oldest pages hoard the internal signal and every new page starts cold. Editing two older pages to link at the new one is the fix, and it takes ten minutes.

   SIGNAL New pages with no inbound internal links a month after publishing.

6. 37

   hreflang, if used at all, returns the tag bidirectionally.

   Every language variant must point at every other one, including itself. One-directional hreflang is worse than none because it is an instruction that cannot be resolved and gets discarded.

   SIGNAL No return tags errors in the international targeting report.

7. 38

   HTTPS everywhere, with no mixed content and an automated certificate renewal.

   A certificate that expires on a Saturday takes the whole site out of the results for as long as it takes somebody to notice. Automate the renewal and alert on it rather than relying on a calendar entry.

   SIGNAL Mixed-content warnings in the console, or HTTP URLs still returning a 200.

The internal-links item sits in this group rather than in the crawl group on purpose. Discovery is the crawl question. This is the editorial one: which pages you chose to link, with what anchor, in both directions. It is the check most likely to be quietly dropped when a team gets busy, and it is the one with the clearest compounding return, which is why it is built into the draft stage of our content automation pipeline rather than left as a quarterly cleanup task.

Structured data closes the list because it is the most over-invested item on it. Ship valid markup for the type the page actually is, then stop. It makes a page eligible for richer presentation and removes ambiguity about the author and the dates. It is not a ranking factor and no amount of additional schema types will make it one. The adjacent work that does pay off on answer surfaces is a content question rather than a markup one, covered in our guide to answer engine optimization.

A worked example: one audit, start to finish.

Take a fictional inventory tool, Stockline. Replatformed seven months ago, 340 real pages, organic traffic down 60% since the migration and flat ever since. Nobody can say what broke. Here is what running the 38 checks in order actually returns, and how long each group took. The company is invented. The distribution of failures is not: this is the standard shape of a post-migration audit.

Five and a half hours to diagnose, most of it in group two, and the finding was a template default that nobody chose. That is the normal outcome. Technical debt in SEO is not usually the product of bad decisions, it is the product of platform defaults surviving a migration that everyone assumed had been checked.

The part worth copying is the order. Groups three through six took three hours and produced real findings, and none of them explained the traffic drop. Had the audit started with performance, which is what most audits do, the two hours of indexation work that held the actual answer would have come last. If you would rather not run this yourself, the done-for-you SEO audit service runs the same pass and returns the findings ranked by what they recover rather than by how many URLs they touch.

Common mistakes in a technical SEO audit.

1. Starting with performance. It is the easiest layer to measure and the most satisfying to report, so it absorbs the attention first. On a page that is well built and not ranking, vitals are usually already passing and the failure is two groups upstream in indexation.
2. Auditing pages instead of templates. Almost every finding in the markup and rendering groups is a template bug repeated across hundreds of URLs. Sampling ten random pages finds the same issue ten times and tells you nothing about how many templates you actually have.
3. Shipping a 400-row issue list. A crawl tool will happily emit every deviation it can detect, sorted by count. Sorted by count is the wrong order: two canonical fixes can outrank 300 alt-text warnings, and a list nobody can prioritise is a list nobody implements.
4. Treating disqualifiers as levers. Fixing redirects and shipping schema are both worth doing and neither will lift a page that nothing links to and nobody searches for. Teams clear the technical layer, see no movement, and conclude the channel does not work.
5. Auditing once, after the damage. The failures that cost the most are introduced at a replatform or a redesign. A pass before launch and another 48 hours after catches them in the window where fixing them is still cheap.
6. Confusing a clean audit with a plan. A site can pass all 38 checks and earn nothing, because passing means the pages are eligible rather than good. When the audit comes back clean, that is the signal to stop auditing and go work on what the pages are for.