I Audited My Site for AI — Here's What I Found
We build an AI-readiness scoring engine for a living, so we do the obvious thing regularly: we point it at our own properties. This post walks through what a full audit of one of our own sites surfaced — including the findings that made us wince — because the gaps we found are, almost line for line, the same gaps we see in the vast majority of business sites we score. Consider this a preview of what your own audit would probably say.
How the audit works
Our scorer grades a site from 0 to 100 across six categories of AI-citation readiness: crawl access and technical parsability, content evidence signals, structured data, passage structure, entity clarity, and off-site corroboration. The evidence-signal category is weighted around the three factors that the Princeton-led GEO study (KDD 2024) validated experimentally — statistics, quotations, and source citations — which improved generative-engine visibility by up to 40% in the researchers' benchmarks.
The uncomfortable premise of any honest audit: what humans experience as a good website and what an AI engine experiences as a citable source are different things. A site can be beautiful, fast, and converting well while being, from a machine's perspective, an anonymous wall of unverifiable prose.
Finding 1: Pages that explained without evidencing
The first thing the scorer flagged on our own content was low evidence density on several pages. The writing was clear and accurate — and almost entirely assertion. Few attributed statistics. No quotable, attributed statements. Sparse citations to external sources. Exactly the pattern we lecture clients about.
Why it matters: an AI engine composing an answer needs sentences it can attribute. "This approach works well" is something the model can generate itself; "researchers measured up to a 40% visibility improvement" with a source attached is something it has to cite. Pages built entirely from the first kind of sentence give engines nothing to take. The fix was mechanical once we saw it: each priority page got specific numbers with attribution, at least one quotable statement from a named source, and links to the research behind claims.
Finding 2: Structured-data gaps in embarrassing places
Our core pages carried Organization schema — but the audit found newer pages that had shipped with none at all, and older ones whose descriptions had drifted out of sync with how we describe the business elsewhere. Individually trivial; collectively the kind of inconsistency that makes an engine less certain about who you are. Machines do not extend the benefit of the doubt. If your homepage schema, your about page, and your directory listings describe you three slightly different ways, you have three weak identities instead of one strong one.
The audit finding that changes how people think is never the exotic one. It's realizing your site answers "who are you and why should I trust this?" with a shrug — on exactly the pages you most need cited.—The ClickRadius team
Finding 3: Content structured for scrollers, not retrievers
Several pages buried their most citable material mid-page under vague headings. Humans scroll; retrieval systems chunk pages into passages and fetch the passage that answers the question. A crisp answer hiding under a heading like "More thoughts" is a passage that loses to a competitor's page where the same answer sits under a heading that states the question. The fix: self-contained sections, each answering one question, with headings that say what the section answers — plus an FAQ where the questions match how buyers actually phrase them.
One detail from this pass surprised us: the worst-structured pages were often our best-written ones. Long, flowing essays read beautifully to humans and atomize terribly for machines — the key claim in paragraph nine of an unbroken narrative is effectively invisible to a retriever chunking by section. The pages that performed best in passage terms were the ones a writing teacher would call choppy: one question, one heading, one self-contained answer, next. We did not make the essays worse; we added structural signposts so both audiences could navigate them.
Finding 4: The off-site footprint was thinner than assumed
The audit's off-site category checks how well independent surfaces corroborate your entity — directories, profiles, third-party mentions. Ours was thinner than we had assumed, which stung, because industry data suggests the majority of what drives citation outcomes lives off-site. This is the finding we see most universally in the sites we score: businesses assume their web presence is what is on their domain. To an AI engine cross-referencing claims, your presence is everything the rest of the web says about you — and for most businesses that is very little.
What we fixed first, and what moved
- Access and identity first: AI-crawler permissions verified, schema completed and made consistent everywhere.
- Evidence into the top pages: statistics, quotations, and cited sources added to the commercially important pages — not the whole site at once.
- Passage restructuring: question-shaped headings, self-contained sections, FAQs.
- Off-site program started: the slow, compounding layer — directories and profiles brought into consistency.
We will not pretend one audit turned into overnight citations — that is not how the mechanism works, and anyone telling you otherwise is selling too hard. What we can say honestly: the score improvements were largest in exactly the categories where the work was mechanical (evidence density, schema, structure), and those are also the categories the research says move citation likelihood. We have watched a site we optimized climb from a readiness score of 45 to 97 by grinding through this same list. The gaps are ordinary, and finding them is the cheap part. The fix is a checklist, not a mystery. And according to every industry analysis of AI-search presence, the large majority of your competitors have not run any audit at all.
Frequently asked questions
What does an AI-readiness audit actually check?
A thorough audit covers six broad areas: whether AI crawlers can access and parse your pages; whether your content carries the citation signals research has validated (statistics, quotations, source citations); whether structured data declares who you are; whether your pages are organized into answerable passages; how your entity is corroborated off-site; and whether AI engines currently mention you at all.
Can I audit my own site without tools?
Partially. You can manually check robots.txt for AI crawler rules, validate your schema with free validators, count statistics and quotations on key pages, and ask the five major engines your buyers' questions. What is hard to do manually is scoring consistently, covering every page, and monitoring changes over time — which is where automated scoring earns its keep.
What should I fix first after an audit?
Fix access and identity first — AI crawler permissions and Organization schema — because nothing else matters if engines cannot fetch or resolve you. Then add evidence (statistics, quotable statements, cited sources) to your two or three most commercially important pages. Off-site entity building runs in parallel as the long-term compounding layer.
Curious what an audit would find on your site? Get your free AI Readiness Score — the same six-category grading we ran on ourselves — or see plans and pricing.