Module 5 · Lesson 2/3 · ~12 min read
0/21 lessons read · Your progress is kept in this browser only.
Finding Inconsistent Brand Facts
A five-step audit to find where your brand facts conflict across the web. Branded questions help you check accuracy; they don't measure visibility.
Learning goals
- Runs a five-step brand-fact audit.
- Builds a table comparing the fact sheet with where facts are published, and marks the conflicts.
- Explains that branded questions are for checking accuracy, not for measuring visibility (link to Module 2).
- Classifies the gaps it finds and decides the order of fixes.
In this lesson
Lesson 1 produced a brand fact sheet. Now we hold it up against the rest of the web: where are your facts written differently, what is the AI actually saying, and what should you fix first?
Step 1: Keep the sheet ready
An audit starts from the sheet. Without one you can't decide what counts as "wrong". At minimum, the sheet needs name, category, address, contact details, hours, pricing and key terms, founding year and founder.
Step 2: Take an inventory
List every place your facts appear: your own site's pages, business and map profiles, review-platform and marketplace listings, social profiles, your replies in community threads, press coverage, partner pages.
Next to each one, note who controls it: you, or someone else.
Step 3: Build a comparison table
Facts go in the rows and places in the columns. Mark a cell "same" when it matches the sheet; otherwise copy what it actually says.
Representative example: Acme Helpdesk is fictional and the values were written for this lesson.
| Fact | Fact sheet | Website | G2 listing | App marketplace listing | Old press article |
|---|---|---|---|---|---|
| Category | Help-desk software for small teams | same | "CRM software" | same | same |
| Starting price | $15 per agent/month | Pricing page: same; 2024 blog post: $12 | $12 | — | — |
| Free trial | 14 days | same | 30 days | same | — |
| Founded | 2018 | About page: 2019 | — | — | same |
The bold cells are conflicts, five in all. The review-platform listing still shows an old category, an old price and an old trial length. A 2024 blog post repeats the old price. And the website disagrees with itself about the founding year.
Step 4: What does the AI say? Use branded questions
Remember Module 2: visibility is measured with non-branded questions; accuracy is about what's said about you. Branded questions inflate visibility, but they're the right tool for accuracy, because here you want the AI to talk about you. You're not asking "does it find me?" You're asking "does it describe me correctly?"
Write one branded question for each important fact on the sheet. For example:
"How much does Acme Helpdesk cost?" "How long is Acme Helpdesk's free trial?" "When was Acme Helpdesk founded?"
These questions were written for this lesson only.
Representative example. The answer is fictional, not real engine output:
"Acme Helpdesk is a CRM tool for small businesses. Plans start at $12 per agent per month, with a 30-day free trial."
All three errors match the same column: the review-platform listing. That's a strong hint about where to start.
Two cautions:
- One answer misleads. As Module 2 showed, the same question can come back differently on another day or another engine. Ask each question several times on different days, and try more than one assistant if you can.
- Keep a record. Log the date, engine, question and answer for each try. A screenshot without date and engine is useless a week later.
Don't mix branded-question results into visibility figures. "The AI named us in 10 of 10 branded questions" isn't a visibility win, because you put your name in the question.
Step 5: Classify the gaps and fix them in order
Ask one question of each conflict:
- Do you control the place? Your website, your vendor profile on a review platform, your marketplace listing and your social profiles all count. Fix these now. For Acme Helpdesk that means the About page, the 2024 blog post (update it or add a dated note), and the G2 listing's category, price and trial.
- Does someone else own it? Press articles and partner pages, for example. Send the publisher a short correction request with a source. If nothing changes, make sure the correct fact is clear and current on your own site.
- Is the error only in the AI answer? Look for where it came from. If the answer shows sources (Module 3), start there. If not, the outdated entries in your table are the likeliest candidates.
Don't expect the AI answer to change right after you fix things. Answers depend on training data and on pages being re-crawled, and there's no fixed timetable for either. Re-ask the same branded questions a few weeks later, again on different days.
Accuracy as a metric
The methodology has a name for this: Brand Accuracy. It shows whether what an answer says about the brand matches the brand facts the customer has verified. Without verified brand facts, no accuracy score is shown.
This lesson explains the definition. Entering facts in the Brand Facts panel on paid plans doesn't by itself produce an accuracy score.
So measuring accuracy starts with a verified fact list like the one from Lesson 1. Without it, "the AI gets us wrong" stays an impression.
Definition: Brand Accuracy.
Summary
- The audit has five steps: fact sheet, inventory, comparison table, branded questions, then classifying and fixing.
- The comparison table maps facts against places. Conflicts tend to cluster in old listings and pages nobody updated.
- Branded questions check accuracy, not visibility, and their results stay out of visibility figures.
- One answer misleads. Ask several times on different days, and log the date and engine.
- Fix the places you control first, then ask others for corrections. There's no fixed timetable for AI answers to change.