What Should an Entity Optimization Audit Check Beyond Website Content?

What should the master fact sheet contain?

Start the audit by writing down the facts every record should agree on:

• legal name, trading name and former names;
• founding date and founders;
• headquarters and registered office, labelled separately;
• leadership, with titles and start dates;
• category description in one sentence;
• products and services, with current names;
• locations, hours and phone numbers for any customer-facing sites;
• identifiers: company registration number, tax or VAT number where public, D-U-N-S number, Legal Entity Identifier if held, and knowledge base IDs.

Confirm it with the client before checking anything else. An audit against the wrong facts sends everyone chasing the wrong fixes.

Which records should the audit check?

Google's knowledge panels can be claimed by the subject or an official representative, who can then suggest changes. Record who in the organisation controls each profile. Unclaimed or orphaned profiles are a common source of stale facts.

What access check do most audits miss?

Whether AI assistants can read each source at all. Robots rules for AI crawlers vary widely between the sites where entity facts live, for example:

• Crunchbase explicitly allows the search crawlers and user-triggered fetchers of OpenAI, Anthropic and Perplexity.
• Yelp disallows those same search crawlers from everything except its articles.
• Instagram blocks OpenAI's, Anthropic's and Perplexity's crawlers, and every crawler it does not list.
• Dun & Bradstreet blocks OpenAI's training crawler from its business directory pages but not other crawlers.

Robots rules are only part of the picture. Some sites that block AI crawlers still reach answers through data licences and partnerships: Yelp blocks OpenAI's and Perplexity's search crawlers on business pages but licenses its business details and reviews to OpenAI and supplies Perplexity, and Google's AI features read whatever Googlebot can. So for each record note the robots rule, any known data partnership, and whether the source is actually cited, and prioritise sources that are cited or licensed. Check the client's own site the same way. How that works is covered in how to audit your website for AI visibility gaps using its URL.

How do you check for name collisions?

Search for the organisation's name in knowledge bases, company registers and the assistants themselves. Record every other entity with the same or a similar name: companies in other countries or industries, products, people and places. For each, note what distinguishes them: location, industry, founding date, website. Collisions explain many wrong answers that look like data errors, such as another company's lawsuit or founder attached to the client. The fix is disambiguation on every record the client controls. Use the full name with a descriptor, the location, and identifiers that link records together.

What should the audit output look like?

A discrepancy matrix, with facts down the side and sources across the top:

Add an access column, showing whether AI search crawlers can read each source, and an owner column showing who controls it. Then sort the fixes. Sources that are wrong, readable by assistants and cited in answers come first. Sources that are wrong but blocked, or rarely cited, come later. How to decide what goes into the report sent to a client is covered in what an agency should check before sending a client an AI visibility report.

How do you tie the audit to what assistants say?

Run a short set of entity questions before and after the fixes, several times each, in the assistants the client's customers use:

• "What is [organisation]?"
• "Where is [organisation] based?"
• "Who runs [organisation]?"
• "How big is [organisation]?"
• "Is [organisation] the same as [similarly named entity]?"

Record each fact given and each source cited, then mark it against the matrix. A wrong fact that matches a specific record tells you which fix matters most. A wrong fact that matches no record points to model knowledge or a collision, which calls for disambiguation and patience rather than profile edits. Repeating the same questions after the fixes gives the client evidence that the audit changed something, not just a list of corrected profiles.

How do you connect the records?

• Organisation markup on the website, with sameAs links to the official profiles and knowledge base entries.
• Identifiers where supported: Google's organisation markup accepts identifiers such as a D-U-N-S number, and encourages the ISO 6523 code field for it.
• Links from profiles back to the website, so each record points to the source of truth.
• Knowledge base identifiers, such as a Crunchbase ID recorded in Wikidata where an item exists.

What might an illustrative case look like?

Consider an illustrative industrial equipment maker whose AI answers gave an old headquarters, a founder as CEO and "about 30 employees" when it had 140. The website was correct. The audit found the old city on LinkedIn and Crunchbase, both readable by AI search crawlers and both cited in answers. The founder was still listed as CEO on three profiles, and the employee count came from an outdated industry database. A trade directory, blocked to AI crawlers, also had old details but was never cited. The company fixed the cited, readable sources first, then the rest over two months. A month after the first fixes, most answers gave the correct headquarters, CEO and size.

What do people ask about entity audits?

How often should an entity audit be repeated?

Yearly as a full audit, and after any change to name, leadership, headquarters or products.

Can the audit be automated?

Parts of it, such as crawler access checks and some profile lookups. Judging which facts are right, and which sources matter, needs a person.

Who should own the fact sheet?

One person in marketing or communications, with a named contact for legal facts. Every profile update should start from it.