AboutWorkServicesToolsBlogContact Get in touch →
SEO

The 10 AI Visibility Mistakes I Fix First on Every Audit

Same ten checks, same order, on every client site I audit. Almost none of them are exotic, they're boring, fixable mistakes that quietly cap how often AI search ever cites the page.

I run the same audit on every new client site before I touch a single page. Same ten checks, same order, every time. I'm sharing the list because the pattern surprised me: it's almost never the exotic stuff. It's ten boring, fixable mistakes that quietly cap how often ChatGPT, Perplexity, and Google AI Overviews ever mention the site at all.

This is the actual checklist, not a summary of it. If you want the full pre-launch version I run before anything ships, I already wrote that one up in the pre-launch SEO/AIO/GEO/AEO/SXO/LLMO checklist. This post is the audit I run on sites that are already live and already struggling to get cited.

Every mistake below fails one of those three gates. If the page can't be crawled, nothing else matters. If the source can't be trusted, the content never gets cited even if it's crawled. If the answer isn't extractable, the model skips it for a competitor who made the answer easier to lift.

The ten mistakes, in audit order

I check these in this exact order because each one builds on the last. There's no point fixing FAQ formatting on a page Google can't crawl yet.

  1. Blocked or slow-crawling pages. I've opened robots.txt files that accidentally block an entire /blog/ directory, left over from a staging config nobody removed. Check robots.txt and the sitemap first, every time, before touching copy.
  2. No sitemap submission or a stale one. A sitemap that still lists pages deleted eight months ago tells crawlers the feed isn't maintained. I regenerate it and resubmit through Search Console and Bing Webmaster Tools on every audit.
  3. Inconsistent entity name across the web. "John Arce Marketing" on the site, "JArce Consulting" on LinkedIn, a third variant in an old press mention. Models resolve who's talking before deciding whether to trust what's said. Pick one name, one spelling, and make it match everywhere.
  4. Missing or broken schema. Markup copied from a template two years ago, never run through a validator, now throwing errors. Broken schema is worse than no schema. It signals a page nobody maintains, and I check it with the Rich Results Test before anything else on this list.
  5. No author or Person entity at all. Pages with zero byline, zero author schema, zero way for a model to attach a name to the claims. I add Person schema with a consistent sameAs list on every page that makes a factual claim.
  6. Burying the answer under three paragraphs of throat-clearing. "In this article, we'll explore..." followed by two paragraphs of context before the actual answer. Most AI summarizers never read that far. I move the direct answer to the first two sentences under a question-shaped heading, every time.
  7. FAQ sections with generic questions and vague answers. "What is SEO?" with a one-line non-answer doesn't help anyone, human or model. I write the actual questions real users ask, with specific, concrete answers, no filler.
  8. Comparisons written as prose instead of tables. A paragraph comparing three tools is harder for an AI summarizer to extract cleanly than the same three tools in a table with consistent columns. I convert every real comparison into a table.
  9. No visible or accurate freshness signal. A page claiming to be "updated 2026" that was last actually touched in 2023, with dateModified in schema that doesn't match. AI systems weight recency hard on anything time-sensitive, and a mismatched date reads as a lie, not a feature.
  10. Isolated pages with zero internal links. A good page that nothing else on the site links to. I've seen pages with real substance get skipped for citation purely because nothing in the topic cluster pointed to them. I cover the linking mechanics in more depth in SEO in the age of super intelligence.
⚠️
The order matters more than the list

I used to run these roughly top-to-bottom but inconsistently. Now I don't move past a gate until it passes. Fixing FAQ copy on a page robots.txt is blocking is wasted work, the model never sees the page to read the FAQ. Crawl, then trust, then extraction, in that order, every audit.

What each mistake actually costs

I don't have access to a universal "AI visibility score," and I'm skeptical of any tool that claims to hand you one with confidence. What I can tell you is which gate each mistake blocks, and roughly how much audit time it costs to fix, because that's the number that actually decides what I tackle first on a client call.

MistakeGate it blocksTypical fix time
Blocked or slow crawlCrawlUnder 30 minutes once found
Stale sitemapCrawlUnder 30 minutes
Inconsistent entity nameTrustA few hours, across every profile
Broken schemaTrust1–2 hours per template
No Person entityTrust1–2 hours, one-time setup
Buried answerExtraction15–30 minutes per page
Weak FAQExtraction30–60 minutes per page
Prose instead of tablesExtraction15–30 minutes per comparison
Bad freshness signalTrustUnder 15 minutes per page
Orphan pagesExtraction30–60 minutes per cluster

The pattern that jumps out: the crawl and trust fixes are mostly cheap and mostly one-time. The extraction fixes are cheap individually but repeat per page. That's why I always clear crawl and trust issues first on a new client, they're the fastest wins, and extraction work only pays off once a model can actually reach and trust the page in the first place.

🔧
The fix I do first on almost every audit

If I only have an hour on a new client site, I spend it on entity consistency. Mismatched names across the site, LinkedIn, and old directory listings is the single most common mistake I find, and it's invisible to the client because nothing looks broken. The page renders fine. It just never gets cited, because the model can't confirm who's behind it.

How I verify a fix actually worked

Fixing the mistake isn't the finish line. I check three things after every pass: Rich Results Test for schema validity, Search Console's Generative AI performance report for impression changes, and a manual spot-check inside ChatGPT and Perplexity asking the exact question the page is supposed to answer. If none of those three move after a fix, I didn't actually fix the thing that was blocking citation, I fixed something else that looked related.

I wrote the longer version of this tracking setup already, including the exact GA4 channel group I use to separate AI referral traffic from regular organic, in SEO in the age of super intelligence. This audit is the upstream half of that process, find and fix the mistakes, then track whether the fix moved anything real.

Key takeaways

  • Ten mistakes repeat across almost every audit, and none of them are exotic. They're boring, and that's exactly why they get missed.
  • Fix in gate order: crawl, then trust, then extraction. Extraction fixes on an uncrawled page are wasted work.
  • Entity consistency is the highest-leverage, lowest-visibility fix. Nothing looks broken, but it quietly caps citation.
  • Broken schema is worse than no schema at all, it signals an unmaintained page.
  • A fix isn't verified until you check it in three places: schema validator, Search Console, and a manual query inside the AI surface itself.

What I'll do next

I'm turning this list into the actual intake checklist I run on the first call with a new client, instead of carrying it in my head. Next audit, I'm also timing each of the ten checks properly instead of estimating, so the fix-time table above gets replaced with real numbers from real audits instead of my rough sense of how long each one takes.

FAQ

What's the very first thing you check on a new client audit? +
Robots.txt and the sitemap. If a page can't be crawled, no amount of content or schema work matters, so I rule that out before looking at anything else.
Is entity consistency really more important than content quality?
It's more important than content quality that nobody can find a reason to trust. A well-written page from a source the model can't confidently identify gets skipped in favor of a mediocre page from a consistent, verifiable entity. Fix both, but fix entity consistency first since it's usually cheaper.
How do I know if my schema is actually broken?
Run the page through Google's Rich Results Test. Copy-pasted schema from an old template is the most common source of errors I find, usually a required field left blank or a type that no longer matches the content on the page.
Do I need to fix all ten mistakes before I see any results?
No. Crawl and trust fixes tend to show up fastest because they unblock everything downstream. Extraction fixes, like FAQ formatting and answer-first structure, compound more slowly but keep paying off as you apply them page by page.
How often should I re-run this audit?
I re-run it on every new client intake, then again roughly every quarter on sites I manage, since sitemaps go stale and schema drifts as pages get edited. A lighter version, just the crawl and freshness checks, is worth running monthly if you're publishing often.
Share this

Want this built for your team?

I design AI agents and growth automation that run without babysitting. If that sounds useful, let's talk.

Get in touch →

Keep reading