The website truth audit: find where your site describes the old company

Every company I have audited carries at least three versions of itself online at once. The homepage says what you are now. A services PDF from 2022 says what you were then. And a forgotten team page, an old case study, or the site's own structured data says something in between. Each surface was true when someone wrote it. Together they disagree, and nobody owns the disagreement.

Humans mostly never notice, because humans only read the pages you send them to. Machines read everything. Search engines and AI systems crawl the stale PDF next to the fresh homepage, and Google is explicit that its AI search features work from the same crawled content as classic search. When your surfaces disagree, you have delegated the choice of which version to repeat.

What is a website truth audit?

A website truth audit is a structured inventory of every public claim your presence makes, and a reconciliation of the places they disagree. The unit of work is the claim, a checkable statement: what you do, who you serve, what you charge for, what you have built, who runs the company, where you operate. The audit finds each claim, records where it lives, and asks the question Google's own content self-assessment puts bluntly: does the content have easily-verified factual errors? Stale claims are exactly that, factual errors with a publication date.

Inventory the surfaces machines actually read

The audit fails if it only covers pages in the navigation. The claim-bearing surfaces include:

• Every indexed page, including the ones no menu links to
• PDFs: brochures, spec sheets, case studies, investor decks
• Page titles and meta descriptions, which often lag a rebrand by years
• Structured data: Organization markup, product markup, FAQ markup
• Feeds and sitemaps
• Machine-facing files, if you publish them: llms.txt (a plain-text site description aimed at AI systems) and agent cards (machine-readable identity files)
• Off-site profiles you control: directories, social bios, app stores

The machine layer deserves its own pass because it is invisible in a browser and highly legible to systems. Google says Organization structured data helps it “disambiguate your organization in search results” and recommends the same name and identity properties everywhere, with sameAs links whose whole job is to unambiguously indicate identity. Markup carrying your old positioning is a claim, published to readers that never expire it.

And those readers matter commercially. One 2026 preprint tracking 100+ brands found roughly 78% of AI-answer citations pointing at corporate websites, a vendor figure to hold loosely, but the direction is the point: your own surfaces are the evidence base AI works from, including the surfaces you forgot you published.

The claim register

Record every claim in the same schema. This is the audit's spine, and a spreadsheet is enough:

Two columns do the most work. Audience prevents false alarms: an investor page and a product page describing the company differently is often correct, and the register records it as audience-specific rather than contradictory. Approval finds the quiet problem: claims nobody owns. In my experience the “approval: unknown” rows outnumber the outright contradictions, and they are where the next stale claim comes from.

Resolve, then design

Group the register by entity and offer, and the contradictions surface on their own: two revenue descriptions, three category labels, a retired product still sold on page 40 of a PDF. Each contradiction gets a decision, made by the person who owns that truth, recorded with a date: which version is correct, what replaces the others, and which surfaces change. That decision log is the real deliverable. Design can start once the log is empty, because now there is an agreed reality to design toward.

One contradiction, resolved, with made-up details so the shape is visible. The pricing page says “serving mid-market teams”; a 2023 case study PDF says “built for enterprise”; the Organization markup still carries the pre-rebrand company name. The owner of positioning decides: mid-market is current truth. The register logs the decision and its date, the PDF gets a corrected replacement at the same URL, the markup gets the current name, and all three rows flip to current. Fifteen minutes of deciding, three surfaces changed, one era of confusion closed.

Resist the shortcut of fixing contradictions silently as you find them. Unrecorded fixes recreate the original problem: claims changed by whoever touched them last, on no one's authority. The register is the difference between cleaning the site and governing it.

What the audit cannot do: purge the copies you do not control. Old press coverage, directory listings, and third-party descriptions keep telling the old story until off-site cleanup and fresh evidence overtake them. Route that work to PR and partnerships, and let the audit stay focused on the surfaces you own.

How I run this

I ran this audit on my own site in 2026, and the artifacts are public by design: the claims on bobmichaels.ai trace to a written record of what the site may claim, and the machine surfaces (metadata, Person and ProfessionalService markup, llms.txt, the agent card) carry the same positioning as the visible pages, with the limits stated instead of smoothed over. One example: the site's machine endpoint is described everywhere as discoverable and callable, and nothing more, because formal protocol conformance has not been independently tested. That is what a resolved claim register looks like in production: strong claims and stated limits that match on every surface.

For clients, this audit is the first stage of a web presence transformation: inventory, claim register, contradiction log, decisions, and then, only then, the rebuild. If your website has been through three positionings and nobody can say which pages carry which era, the audit is where the untangling starts.

Questions worth asking next

Sources

1. Google Search Central, creating helpful, reliable, people-first content (self-assessment questions; E-E-A-T). https://developers.google.com/search/docs/fundamentals/creating-helpful-content
2. Google Search Central, Organization structured data (disambiguation, consistent identity properties). https://developers.google.com/search/docs/appearance/structured-data/organization
3. Schema.org, Organization type (sameAs and entity identity). https://schema.org/Organization
4. Google Search Central, "AI features and your website" guidance. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
5. Pratyush Kumar (Ranqo), "Generative Engine Optimization at Scale," arXiv preprint 2606.20065, June 2026. https://arxiv.org/abs/2606.20065