When a customer asks ChatGPT your closing time, the AI doesn't check your website in real time. It draws on whatever it absorbed during training — and if your Yelp page, your Google Business Profile, and your own site all say something different, the AI has to make a judgment call. It often gets it wrong.
Why AI Sees Multiple Versions of Your Business
Every time your business has been listed somewhere online — Yelp, TripAdvisor, a local chamber directory, a neighborhood blog post from 2021 — that source contributed to what AI models learned about you. AI training data isn't a single clean feed. It's a messy pile of overlapping snapshots taken at different times by different crawlers.
A bakery in Phoenix might have:
- A Google Business Profile showing hours of 7am–3pm (updated last month)
- A Yelp listing still showing 7am–5pm from before the owner cut afternoon hours
- A food blogger's "best brunch spots" post from 2023 listing 7am–2pm
- The bakery's own website showing 8am–3pm after a recent rebrand
All four sources are real. All four disagree.
What AI Does With Conflicting Data
AI models don't vote on facts the way a search algorithm calculates a consensus. They weight sources by perceived authority and recency — but imperfectly, because training snapshots happen months before a model ships, and that process doesn't cleanly resolve contradictions.
In practice, the AI tends to do one of three things.
Defer to high-authority sources. Google Business Profiles carry significant weight because they're referenced across billions of web pages. That signal is strong — but not always stronger than a dozen blog posts all repeating an old piece of information.
Hedge. You've probably seen AI answer "hours may vary — I'd recommend calling ahead." That's often a conflict hedge: the model detected inconsistency and retreated to a non-answer rather than commit to one version.
Pick the most repeated claim. If eight directories all copied the same outdated entry from a 2019 data feed, the AI may treat that repetition as evidence of truth. This is how a phone number you changed five years ago keeps showing up in AI answers.
The Aggregator Copy Problem
Business directories frequently scrape each other. One platform ingests another's data, a third platform ingests both, and the original error multiplies. AI training data picks up all three copies without knowing they're duplicates of the same wrong entry.
A dental practice in Austin found its old address — from before a 2022 office move — appearing in ChatGPT answers. A data aggregator still had the stale record, and it had been republished across dozens of niche directories. The practice had updated Google and Yelp but never touched the upstream feed. The AI wasn't making things up. It accurately reported what the most widely repeated source said.
Which Conflicts Hurt Most
Not all mismatches matter equally. Three categories cause the most customer friction.
Address disagreements. A customer who shows up at the wrong address walks away and often leaves a one-star review. Address errors compound because mapping apps frequently pull from the same stale feeds.
Hours. Conflicts here lead to missed visits. "Showed up and the store was closed" is one of the most common negative reviews for businesses with AI visibility problems.
Service descriptions. If three sources say your electrician handles residential work and one outdated press release describes you as commercial-only, AI may steer the wrong clients away — or confuse the pitch entirely.
How to Reduce Conflicting Signals
You can't control which sources AI models trained on, but you can shrink the surface area of disagreement going forward.
Audit the major aggregators first. Four US data aggregators — Data Axle, Localeze, Foursquare, and Acxiom — push records to hundreds of downstream directories. Fixing your entry at the source is more effective than chasing individual listings. Services like Yext or Whitespark can push corrections in bulk, though propagation takes weeks.
Make your own website the canonical source. Use LocalBusiness schema to mark up your hours, address, phone number, and services with structured data. When AI crawlers hit your site, the schema signals "this is the authoritative record." Include those same details in plain text on the page too — AI can't always read information buried in JavaScript or non-semantic markup.
Add a llms.txt file. This plain-text file tells AI crawlers what matters on your site. A few lines pointing to your contact page, hours, and service list nudge an AI toward your version of the facts rather than an aggregator's stale copy. The llms.txt guide for local businesses shows the exact format to use.
Date your content. An "Updated September 2026" note near your hours or address tells both search engines and AI that this version is current. An undated page looks the same as a page from 2019 — the AI has no way to weight it higher.
Start by Seeing What AI Is Actually Saying
Before you can fix anything, you need to know what's broken. Run a free scan to see the grade AI assistants give your business — it tests how accurately they answer the core questions customers ask about your hours, location, services, and contact details. If your score is a C or lower, conflicting directory data is usually part of the diagnosis.
Work through one source at a time: Google Business Profile first, then Data Axle, then your website's schema markup. Consistency doesn't have to be perfect everywhere — it has to be right in the places that get cited most.