Prepare the audit and collect the data
A useful audit starts with a clear scope and reliable data. Before touching a single crawl, you define the objective: recover lost traffic, prepare a redesign, or diagnose stagnation. The objective dictates the priorities.
Then gather three sources. Google Search Console gives the real indexing, impression and click data. A crawler like Screaming Frog reproduces Googlebot's journey across every URL. An authority tool like Ahrefs or Semrush exposes the link profile and ranking keywords. Without these three angles, you work blind.
Frame the scope
List the URLs to audit, the page templates (homepage, category, product, article), and the languages if the site is multilingual. Note the reference version: desktop or mobile. Google has indexed mobile-first for years, so your analysis starts from the mobile version.
Finally, export a comparison baseline: current rankings on your target queries, organic traffic over the last twelve months, most-visited pages. These figures will be used to measure the effect of the fixes. An audit that isn't measured can't be defended.
Technical audit: crawl, indexing, rendering
The technical layer determines whether your pages exist in the eyes of the engines. If Google can't crawl, render and index a page, no content work matters. So you start there.
Crawlability and indexing
Run the crawler on the full site, then cross-reference with Search Console. Look for gaps: pages crawled but not indexed, orphan pages missing from the link graph, redirect chains, 404 and 5xx errors. Check the robots.txt line by line and verify that the XML sitemap lists only canonical URLs returning 200.
Canonical tags deserve particular attention. A canonical that points to a URL other than the page itself, across hundreds of products, can de-index an entire catalog. It's the most costly silent error I encounter in audits.
Rendering and performance
Check how each page template renders. A site that loads its main content via client-side JavaScript takes a major risk: LLMs don't execute JavaScript, and Google's rendering remains slower and more fragile than static HTML. SSR or pre-rendered HTML isn't an option, it's the foundation.
Then measure Core Web Vitals across your templates, not just the homepage. LCP, INP and CLS must turn green on mobile. For the detail of the thresholds and fixes, follow our dedicated method for Core Web Vitals.
| Criterion | Client-side rendering (CSR) | Server-side / static rendering (SSR) |
|---|---|---|
| Googlebot access | Deferred, dependent on JS rendering | Immediate, full HTML |
| LLM access (ChatGPT, Perplexity) | Invisible content, no JS execution | Content read directly |
| Indexing risk | High | Low |
| AI citability | Near zero | Optimal |
On-page and content audit
On-page translates search intent into readable signals. A technically perfect but off-topic page won't rank. So you assess relevance before mechanics.
Structure and tags
Check the uniqueness and relevance of title tags and meta descriptions, the presence of a single H1 per page, and an Hn hierarchy with no skipped levels. An H2 should never jump straight to an H4. This structure helps readers and engines alike understand the page outline.
Check internal linking. Every strategic page should receive internal links with descriptive anchors. A service page like an SEO agency gains authority when your relevant blog articles point to it with natural anchors. Distribute the link equity toward what matters commercially.
Intent and content depth
For each key page, ask the question: does it really answer the query? Compare with the current top 10. If competitors cover ten sub-questions and you cover three, the depth deficit explains the ranking.
Even though 92% of AI Overview citations come from the top 10, 47% of positions 5-10 are picked up. Being in the top 10 remains the entry ticket, not just the podium.
Also work on citability. Open each section with a direct, self-contained answer. A passage of 134 to 167 words, factual and complete, maximizes your chances of extraction by an AI Overview or an LLM. It's the format these systems prefer to cite.
Off-site audit and AI citability
Off-site measures the trust the web grants your brand. Content and technical work make you eligible; external authority makes you win. But the nature of that authority changes with AI engines.
Link profile and mentions
Audit the backlink profile: quality of referring domains, anchor diversity, toxic links to disavow. Spot the domains that point to your competitors but not to you: those are your priority link-building targets.
Above all, stop reducing off-site to links. An Ahrefs analysis covering 200,000 domains (December 2025) shows that off-site brand mentions correlate more strongly with AI citations (YouTube at 0.737, Reddit, Wikipedia present in 47.9% of ChatGPT citations) than Domain Rating, which weighs only 0.266. The conclusion is clear: being mentioned counts as much as being linked.
Visibility in generative engines
Check your presence in ChatGPT, Perplexity and AI Overviews on your target queries. It's still open ground: only 11% of domains are cited by both ChatGPT and AI Overviews. Most brands occupy only one ecosystem.
Catalog where your brand is cited outside your site: YouTube, Reddit, Wikipedia, press, industry directories. These signals feed AI citations.
Query ChatGPT and Perplexity on your target queries. Note whether you're cited, by whom, and which sources come up in your place.
Verify that your robots.txt doesn't block AI agents and that your pages serve usable static HTML.
FAQPage schema is a strong signal for AI Overviews. Add it to your high-intent informational pages.
To structure this AI layer step by step, our 2026 SEO audit checklist details every control point, from crawlers to structured data.
Prioritize actions and build the plan
Prioritization turns the diagnosis into results. An audit produces dozens of findings; without a hierarchy, the team handles the easy before the important. So you rank each action along two axes: impact on traffic and remediation effort.
The impact / effort matrix
Place each issue in a matrix. High impact and low effort: do immediately. High impact and high effort: schedule as a project. Low impact and low effort: batch together. Low impact and high effort: drop for now.
A misconfigured canonical across an entire catalog, fixed in one line of template, is the archetype of high impact / low effort. A complete rebuild of the rendering into SSR is a scheduled project. This sorting immediately gives a credible battle plan.
From finding to costed plan
For each priority action, state the issue, the page or template concerned, the precise action, the owner and the expected effect. A deliverable audit always ends with this roadmap, not with a raw list.
Finally, return to the comparison baseline from step 1. Define the success indicators: indexing of target pages, Core Web Vitals turning green, ranking gains, appearance in AI Overviews. Without measurement, you'll never know what worked. To save time on the deliverable structure, grab our GEO Audit Templates Pack: prioritization grids and action-plan templates ready to use.
Request your free GEO audit: we test your target queries in AI engines and hand you a prioritized action plan.
Questions fréquentes
How long does a complete SEO audit take?+
Plan for two to five days for a mid-sized site. Collection and crawling take a few hours, but analysis and prioritization take the most time. A large e-commerce site may require one to two weeks.
Which tools are essential for an SEO audit?+
Three categories are enough: a crawler like Screaming Frog, Google Search Console for real indexing and impression data, and an authority tool like Ahrefs or Semrush. PageSpeed Insights rounds out the performance side.
How often should you redo an SEO audit?+
A complete audit once or twice a year is enough for most sites. Add light monthly monitoring of Core Web Vitals, indexing errors and rankings on your target queries to catch drift early.
Does the SEO audit cover visibility in ChatGPT and AI Overviews?+
Yes, a modern audit includes the AI citability layer. LLMs do not execute JavaScript: static HTML rendering, citable passages of 134 to 167 words and FAQPage schema determine your chances of being cited in generative answers.
