AI Gets Your Phone Number Wrong 36% of the Time

Here's a number that should worry every Ottawa business owner: 36%. That's how often AI chatbots hand out the wrong phone number when somebody asks for a local business. ChatGPT, Perplexity, Gemini, Claude — all of them get it wrong more than a third of the time, according to a 2026 Seer Interactive study.

Now scale that. You spend $400/month on Google Ads. You optimize your Google Business Profile. You chase reviews. Then a prospect asks Siri "what's the number for [your business]?" and gets a five-year-old line that's been disconnected. Or worse — a competitor's.

You can't see it happening. You don't get a notification. You just notice fewer calls than the marketing math says you should be getting. This is the silent version of missed call recovery — the call never even made it to your phone.

So-what: AI-era local SEO isn't about ranking on Google anymore. It's about being cited correctly across half a dozen assistants you don't control.

The 36% Number Isn't a Fluke

Seer Interactive ran the test across the four major models. Phone numbers wrong 36% of the time. Hours wrong about as often. Addresses slightly better, but still in the 20-30% error range.

The deeper finding is uglier. Google Business Profile data — the thing every local SEO agency tells you to obsess over — matches AI output only 27% of the time. Your customer service page matches 64%. AI models that bother to cite a source: 93%.

Read that again. Your GBP is the worst-matching source of all. The thing you've been told is the foundation of local search is the least likely place AI is pulling from. Why? Because most GBPs are stale, inconsistent, or blocked from the AI's training data.

This matters because zero-click searches now make up around 69% of all queries (Similarweb, 2026). When AI gives the wrong number, the searcher never reaches you. They never know they were looking for you. The marketing funnel breaks before anyone clicks anything.

So-what: the canonical version of "where does AI get my info" is not what most agencies are telling you. The data is in different places than you think.

Why AI Hallucinates Your Business Info

LLMs are trained on snapshots of the web. Those snapshots are months or years old. They blend information from your website, Yelp, BBB, an old Yellowpages listing, an industry directory, a competitor's blog mentioning you, and a forum post from 2019.

When those sources disagree — and they almost always do for any business that's been around more than three years — the model picks one. Often the wrong one. The model has no idea which is current. It's pattern-matching, not fact-checking.

The silent killer here is inconsistent NAP — Name, Address, Phone. You moved offices in 2022. You updated your GBP. But the old address still lives on 14 directories, three local press mentions, and your old Facebook page. The AI sees the inconsistency and picks the most "confident" source. Confidence often equals "appears in more places," not "is correct."

Schema markup is the other half. If your homepage doesn't have schema.org/LocalBusiness JSON-LD with your current phone, address, and hours, the AI is guessing from page text. Which means it can pick up an old number from a footer you forgot to update.

So-what: AI doesn't know what's current. It picks what's consistent. Consistency is now an SEO discipline.

The Fix: Become AI's Source of Truth

Three moves. None of them are exciting. All of them work.

First, run a NAP audit on your top 30 citations. Same exact business name. Same exact address format. Same exact phone format (Ottawa businesses: pick one — (613) 555-1234 or 613-555-1234 — and use it everywhere). Update GBP, Yelp, BBB, Apple Maps, Bing Places, your provincial business registry, and your top 5 industry directories. This is the unglamorous work most local SEO agencies skip.

Second, add LocalBusiness JSON-LD to your homepage and contact page. Phone, address, hours, geo-coordinates, opening times. Match GBP exactly. If you can't read JSON-LD, your dev can — it's a 30-minute job.

Third, get cited by trusted Canadian sources. Local press (Ottawa Citizen, Ottawa Business Journal, OBJ.ca), industry .org sites, and any .gov or .edu mention you can earn. AI models heavily weight high-trust domains. One mention in the Ottawa Citizen is worth twenty in random directories.

This is the same hierarchy that makes ChatGPT cite Yelp before your website — and the same fix. Become the most consistent, highest-trust source for your own information.

So-what: AI accuracy is fixable. Most of it is data hygiene, not magic.

What Ottawa SMBs Should Do This Week

Don't take my word for the 36% number. Run the test yourself. Open ChatGPT, Perplexity, Gemini, and Claude. In each one, type: "What is the phone number for [your business name] in Ottawa?"

Then check Apple Maps, Bing, and DuckDuckGo. Five minutes, six surfaces. Write down what each returns.

If even one is wrong, you have an AI citation problem. The fix is the three-step playbook above, plus a quarterly recurring audit (calendar it — most owners forget within six weeks).

One Canada-specific note: wrong contact info in your email footer or website is also a CASL compliance issue. The same NAP audit fixes both problems at once. Two birds, one Saturday morning.

So-what: the leak is silent, the fix is unglamorous, and the businesses that do this work in May 2026 will compound an advantage for the next two years while everyone else chases the latest local SEO trend.