The SEO audit checklist that flags the false positives too.

Skip this phase and you will produce a flat list of pages instead of a diagnosis. Everything category-level depends on work you do before the crawler starts.

• Open 30 real URLs and read the patternIs there folder structure, or does everything sit at root level? Do not take this from the sitemap index alone.
• If URLs are flat, find the category in the DOMThe label is almost always still rendered on the page. Grab its XPath and add it as a custom extraction on the crawl.
• Extract published date the same wayIf it will not resolve cleanly, take the whole containing block and parse the date out afterwards. A messy extraction beats none.
• Build URL groups and join them to Search ConsoleClicks, impressions, CTR and position per group. This is the join that makes every later insight possible.
• Agree the competitor set before you analyse anythingReal businesses competing for the same customer. Exclude directories, marketplaces, news sites and Wikipedia.
• Confirm read-only access to Analytics and Search ConsoleNine indexing reports and the full query data live in Search Console and no crawler can see any of it.
• Check whether a firewall will block your crawlerGet the user agent allowlisted up front rather than discovering a wall of 403s halfway through.
• Note the platform and what it will not let you changeSome findings are platform limits, not mistakes. Knowing which is which stops you writing tickets nobody can action.

The unglamorous foundation. A site can have excellent content and stay invisible because the first two files a crawler looks for are missing or contradict each other.

• Request every sitemap variant directlyDo not assume. /sitemap.xml, /sitemap_index.xml, and whatever robots.txt actually declares.
• Confirm the sitemap only lists indexable, canonical, 200 URLsRedirects, 404s and noindexed URLs in a sitemap waste crawl budget and confuse the picture.
• Read robots.txt line by line, including the AI crawler groupConsecutive user-agent lines form one group. A stack of named bots followed by a wildcard inherits the wildcard rules.
• Check the canonical on every template typeSelf-referencing on the page itself, and pointing where you expect on paginated, filtered and parameter URLs.
• Find noindex directives that should not be thereBoth meta robots and the X-Robots-Tag header. A staging directive that survived launch is a classic.
• Test trailing slash consistencyBoth versions should not resolve with a 200. One canonical form, everything else redirected.
• Check http to https and non-www to www resolve in one hopTwo redirects to reach the homepage is a small thing that shows up on every single crawl.
• Look for parameter and faceted URLs in the indexFilter combinations indexed at scale is the most common cause of index bloat on ecommerce.
• Compare indexed page count against real page countA large gap in either direction is the finding. Too many means bloat, too few means a blocker.
• Check the staging site is not indexableSearch for the staging subdomain. It happens more than anyone admits.

Every internal link should land on a 200. Where it does not, the useful half of the finding is the page containing the bad link, not the broken URL.

• List internal 4xx errors with their source pages"This URL 404s" is not actionable. "This URL 404s and is linked from these six pages" is a ticket.
• Find internal links pointing at redirectsUpdate the link to the destination. Free to fix, and it stops the chains forming in the first place.
• Map redirect chains and loopsAnything over one hop gets flattened to a single 301.
• Find redirects that terminate in a 404 or a noindexed pageThe worst kind, because the redirect looks correct in a report and delivers nothing.
• Test old URLs after any slug changeDo not assume the CMS wrote a redirect. Curl the old URL with a cache-buster and read the actual status code.
• Check for soft 404sPages returning 200 with nothing on them. Search Console flags these and crawlers usually will not.
• Review blocked external resourcesFonts, scripts and images blocked at the source. Compare the rendered page before deciding it matters.
• On migrated sites, sample the old URL list end to endTake 50 URLs from the pre-migration index and confirm each one lands somewhere sensible.

Title length is measured in pixels, not characters, and no tool can tell you what Google is actually doing with your title. For that you go and look.

• Flag titles beyond the pixel thresholdCrawlers default to around 561 pixels, and SERP truncation sits closer to 580 to 600. Treat the flag as "may be trimmed", not "is broken".
• Run a site: query and read what Google rendersGoogle is consistent about which element it drops. Identify it before rewriting anything.
• Check whether the theme appends a suffixIf the brand name is being appended by the template and then trimmed, the fix is the template setting, not 200 rewritten titles.
• Find titles under 30 charactersShort titles are wasted space rather than broken. Worth filling on commercial pages.
• Find duplicate titles across templatesUsually paginated archives or near-identical location pages. Both need distinguishing.
• Check meta descriptions on money pages onlyThey do not rank the page. They do affect whether anyone clicks it, so prioritise by impressions.
• Find pages with multiple H1s or an H1 over 70 charactersRarely fatal, frequently a symptom of a page builder emitting headings for layout.
• Check heading order and duplicationH2s repeated across a template, or a page that jumps from H1 to H4, both make structure unreadable.
• Flag question-format headings on commercial pagesA service page with a question as its H1 is usually chasing the wrong intent.
• Check for thin and duplicated body contentEspecially category descriptions and location pages built from one template.

Two separate jobs here: find what is missing, and throw out the validation errors that do not actually matter.

• Confirm sitewide Organization and WebSite nodes existThe foundation. Everything else hangs off them.
• Map missing schema by page typeThe real gap is usually types absent entirely, not properties that failed validation.
• Re-run every flagged URL through the Rich Results TestOnly escalate what fails there. A schema.org validator warning is not automatically a Google problem.
• Check the graph is linked with @idFloating disconnected blocks are weaker than one connected graph, and duplicate entities confuse it further.
• Look for duplicate schema from two pluginsAn SEO plugin and a theme both emitting Organization is common and worth catching.
• Verify claimed ratings are backed by real review dataStar ratings in markup with nothing behind them is a manual action risk, not an opportunity.
• Match the schema type to what the page actually isService pages are not Products. Pages with no price should not be claiming offers.
• Parse every JSON-LD block programmaticallyA block that will not parse is reported by Google at the first typed field it cannot read, which is rarely where the real fault is.

Two failure modes: equity leaving faster than it circulates, and links carrying no signal because they have no anchor text.

• Find pages with unusually high external outlinksNormal on review and affiliate sites. It only becomes a problem without balancing internal links.
• Find internal links with empty or image-only anchorsA link with no anchor text passes far less about the destination.
• Check anchor text variety per target pageEvery internal link to a page using the identical anchor is its own kind of signal.
• Find orphan pagesIndexable, in the sitemap, and linked from nothing. Crawl plus sitemap comparison surfaces them.
• Check click depth to money pagesIf a commercial page is five clicks from the homepage, the architecture is arguing against it.
• Find tracking parameters inside internal hrefsThey create duplicate URLs and they are almost always accidental.
• Check cross-origin links carry rel protectionA small security and trust item that takes minutes.
• Compare referring domains against total backlinksAn inflated ratio points at sitewide or scraper links rather than genuine coverage.
• Segment nofollow from editorial linksNews, PR and profile links behave differently from earned editorial ones.
• Check country relevance against the target marketA backlink profile from the wrong region is worth knowing about before anyone budgets for more.
• Review the anchor text profile, then decide on disavowDisavow is a last resort, not a routine step. Most profiles do not need one.

Separate what you measured from what you estimated, and say which is which in the report.

• Pull real field data where the site has the traffic for itLab scores presented as user experience is the most common overclaim in this section.
• Identify the LCP element on each templateYou cannot fix largest contentful paint without knowing what it is on that page type.
• Check the LCP image is not lazy loadedLazy loading the hero image actively delays the metric you are trying to improve.
• Find images missing width and height attributesThe cheapest layout shift fix there is.
• List images over 100 KBIndividually minor, meaningful in aggregate, and usually a five minute fix at the template level.
• Check alt text coverageAn accessibility requirement first and an image search opportunity second.
• Review render-blocking resources and total page weightCount the fonts and the weights while you are there. Sprawl is normal and rarely intentional.
• Check Cache-Control on the homepage and key templatesMissing cache headers on a static template is quiet, constant waste.
• Compare the rendered DOM against the raw HTMLIf the content only exists after JavaScript runs, that is the finding.
• Check the security headersHSTS, Content-Security-Policy, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Permissions-Policy.
• Check for server banner disclosure and wrong content typesLow severity, quick to close, and it signals whether anyone is minding the server.

The reports no crawler can see, and the analysis that turns findings into a business case. Report the clean ones too, because that is how a client sees the audit was thorough rather than selective.

• Work through each indexing state separatelyNot found, blocked by 4xx, crawled currently not indexed, discovered currently not indexed, indexable but not indexed. Different states, different causes.
• List URLs with no search analytics data at allIndexed and never returned for a single query is a different problem from ranking badly.
• Check for manual actions and read the indexing trendTrend direction over 12 months tells you whether this is a decline or a plateau.
• Chart CTR by URL group, never sitewide onlyA 2.4% site average can hide one category at 5% and another at 0.5%. Only the second one is the opportunity.
• Sort by impressions to find demand without clicksHigh impressions and low CTR is the shortest path between an audit and revenue.
• Model the upside at 1% through 5% CTRLabel it a model, not a forecast. A number gets budget approved where an adjective does not.
• Age-gate every page before calling it weakA URL with no clicks that is three weeks old needs time, not deletion. Say what window you used.
• Build a year-over-year click loss listWhich URLs lost, how much, and when. That is where the real story usually is.
• Give every weak page one of three decisionsKeep and improve, keep but noindex, or remove and redirect to a named destination. Never leave it open.
• Bucket Google Discover URLs by content typeWhichever type dominates is what the algorithm is willing to amplify. That is a content strategy signal, not a technical issue.

The newest section and the oldest discipline. One is about being retrievable by machines, the other is about whether any of your work gets implemented.

• Check content exists in the initial HTMLToday most assistants fetch raw HTML rather than rendering the page, so content that only appears after JavaScript runs is content they will miss. That may change. Right now, assume it will not.
• Check author bylines are real and connected to schemaAn unattributed page is harder to cite and harder to trust, for readers and machines alike.
• Look for direct-answer formatting and comparison tablesPassages that answer a question cleanly are the passages that get quoted.
• Treat llms.txt as a bet, not a leverAs things stand today, no major AI crawler actively consumes it and Google Search does not use it for ranking. That is where it sits for now. It costs almost nothing to publish and it may well matter later, so add it if you like. Just do not bill it as a ranking factor while it is not one.
• Write every finding as an instruction with a sheet and an exampleThe imperative instruction, the list of affected URLs, and one worked example. Findings without actions do not ship.