This is the single highest-return exercise in AI visibility, it requires no tools, and almost nobody does it.
Set aside half a day. You will need a spreadsheet and a browser.
Step one: define the canonical version
Before looking at anything, write down the correct answer. One name form. One one-line description. One primary category. One address. One phone number.
This sounds trivial and reliably generates an argument. Is it the legal name or the trading name? Is the description the one on the website or the one on LinkedIn? Which of the three category selections is right?
Settle it now, write it down, and give it an owner. Auditing before you have decided the correct answer means doing the work twice.
If two people in the meeting give different one-line descriptions of the business from memory, you have found the root cause before starting the audit.
Step two: search yourself, properly
Search your business name in a clean browser session and read three full pages of results. Not skim — read.
You are looking for every property that describes your business: your own site, social profiles, directories, industry bodies, review platforms, aggregators, old press releases, job boards, supplier registers.
Most organisations find between fifteen and forty properties. A meaningful proportion will contain something wrong.
Step three: build the contradiction sheet
One row per property, six columns.
The contradiction sheet
The right column is what turns an audit into a work list.
Columns to record
- Property name and URL
- Name form used
- Description used
- Category used
- Address and phone shown
- Last updated, if visible
Then mark each row
- Agrees with canonical — no action
- Wrong but editable by us — fix now
- Wrong, needs a claim process — start it
- Wrong, no contact route — note as risk
- Duplicate listing — request merge
- Dead property — request removal
Step four: correct at source
This is the slow part and the part that matters. Each wrong property has to be corrected individually.
Some will have an obvious edit link. Some require claiming the listing, which means a verification process. Some aggregators pull from other aggregators, so fixing the upstream source resolves several downstream ones — worth identifying before working through them one by one.
Publishing a correction on your own site does not fix this. Your site is one input against several agreeing wrong ones, and the wrong version usually wins on weight of agreement.
Step five: anchor it with schema
Once the properties agree, add Organization schema to your site carrying the canonical name, description, address and contact point — plus a sameAs array listing every profile you control.
That array does specific work. It tells a machine that these scattered properties are the same entity as this website, rather than leaving the connection to be inferred. Inference fails more often than people assume, particularly for businesses with common names.
Step six: diary the next one
Entity data decays. Directories change formats, staff update one profile and forget others, an office moves and four listings are missed, an aggregator republishes an old record.
Book the next audit for three months out before you close this one. Treating this as a project rather than maintenance is why most organisations find inconsistencies eighteen months after their last cleanup.
Common mistakes
- Auditing before agreeing the canonical version, then redoing the work.
- Fixing your own site and assuming third parties will follow.
- Skipping properties that require a claim process because they are tedious.
- Using different name forms on social profiles for stylistic reasons.
- Omitting the sameAs array, leaving the connection to inference.
- Treating it as done rather than diarised.