It was a Tuesday in February, around 9:40 AM, and I was drinking a cold coffee at my desk in Markham. A homeowner had just emailed me saying she'd asked ChatGPT for "best kitchen renovation contractor near Markham" and the three names it gave her were not Yellow Pencil. I'd been ranking second on Google for that exact phrase for almost eleven months.
I opened a second browser, asked Perplexity the same question, and got a different three names. Claude gave me two of the ChatGPT names plus a contractor I'd never heard of. Not one of them mentioned me.
That's the moment I realized Google rank and ChatGPT citations are two different universes. And the one my customers were starting to use was the one I was invisible in.
The spreadsheet that started everything
I did what any stubborn contractor does. I opened a Google Sheet and typed 42 queries a real homeowner might ask an AI. "Kitchen remodel cost Markham." "Contractor who does permit drawings Durham Region." "Who renovated the house on [street] in Unionville." Then I ran each query through six models and logged who got cited.
Yellow Pencil showed up in 4 out of 42. Under 10 percent. Meanwhile my Google Search Console was telling me I was impressed on 1,100 queries a month and clicked through at 6.2 percent. My SEO numbers were fine. My AI numbers were terrible.
What bothered me more: the three competitors ChatGPT kept citing weren't ranking above me on Google. Two of them were on page two. One didn't even have a proper website, just a Google Business Profile and a Houzz listing with 180 reviews. That told me something I hadn't wanted to admit.
Ranking and citing are not the same game
Google's job is to hand you ten blue links and let you pick. The AI's job is to pick for you. Those are fundamentally different problems. A blue-link engine rewards on-page SEO: title tags, backlinks, page speed, keyword density. An answer engine rewards something else entirely.
After about six weeks of testing, I'd narrowed the signals that seemed to actually move AI citations into four buckets. I now call them the GEO Score dimensions:
- Entity Readiness — does the model even know you exist as a distinct business? Do you have a Wikidata entry, a consistent NAP across 40+ directories, a schema.org LocalBusiness block that resolves cleanly?
- Answer Coverage — when a real customer asks a real question, does your site contain the actual answer in the actual words? Not "premium craftsmanship." The number. The timeline. The permit fee.
- Evidence Strength — third-party proof. Reviews, press mentions, forum threads, industry citations. The AI doesn't trust you about you. It trusts other people about you.
- Crawlability — can the bots actually fetch your pages? OpenAI's bot, Perplexity's, Anthropic's, Google-Extended. Half the sites I checked were blocking at least one of them in robots.txt without knowing it.
The Houzz contractor with the bad website? He was crushing me on three of those four. Not because he was smart about it. Because his customers had done the Evidence work for him by writing 180 reviews.
What I built for myself first
I'm a contractor, not a data scientist, but I'm comfortable reading logs. I wrote a scraper that ran those 42 queries weekly across eight AI models. Then I scored Yellow Pencil's pages against the four dimensions and logged everything into a dashboard.
In week 3 I fixed the schema on my service pages. Citations moved from 4/42 to 9/42.
In week 5 I added a "typical costs in Markham" section to six pages with actual dollar ranges ($48,000-$92,000 for a mid-size kitchen, which is what I actually charge). Citations went to 14/42.
In week 7 I unblocked GPTBot and PerplexityBot from my robots.txt — they'd been blocked by default in my old WordPress security plugin. Citations went to 19/42.
The moment I saw GPTBot in my access logs for the first time, requesting /kitchen-renovations-markham/, I knew this wasn't a hobby anymore.
That's when two contractor friends asked me to run the same audit on their businesses. And then a dentist. And then a mortgage broker in Oakville. Same story every time — strong Google presence, invisible in AI.
Why I turned it into a product
I kept getting the same question: "Can you do this for me every week?" Which meant the scoring, the tracking, the re-testing as models drift. Because they do drift. Between January and March I watched the same query on the same model return three completely different contractors on three different weeks with nothing changing on any of our sites.
A one-time audit is almost useless in this space. You need weekly re-scoring or you're flying blind.
RankingLocal.ai is what I built to scale the thing I'd been doing manually. It runs the queries, scores the four dimensions, shows you exactly which dimension is hurting you, and writes the findings out in plain English through a persona I call Flare. Flare is an AI advisor but it's trained on what I, a working contractor, actually want to read when I open my dashboard on a Monday morning. Not "optimize your content strategy." Stuff like: "Your robots.txt is blocking Perplexity. Here's the line. Remove it."
The thing I wish somebody had told me
If you run a local service business — contractor, lawyer, dentist, accountant, whatever — your next fifty customers are going to split. Half will Google you. Half will ask an AI. The half that asks the AI will never see your Google rank, no matter how good it is.
And the gap between those two universes is widening, not narrowing. I'm watching it happen in my own logs every week.
You don't need to hire an agency to fix this. Most of it is mechanical. Fix your schema. Unblock the bots. Add specific numbers to your pages. Collect third-party proof. Then monitor it, because it drifts.
What to do if you're reading this
Go to /free-tools/ and run the Robots Check first — it takes 20 seconds and roughly one in three contractor sites I test is blocking at least one AI crawler by accident. Then run the GEO Grader on your homepage to see which of the four dimensions is your weakest. Both are free, no signup.
If you want the full picture — weekly re-scoring across all eight models, query tracking, competitor comparison — the Pro trial is 30 days, no credit card, auto-cancels if you forget about it.
I built this because I needed it for Yellow Pencil. I kept building it because every contractor I showed it to said "I need this too." If you try it and something's broken, or it's missing a feature you'd pay for, hello@rankinglocal.ai is read by me directly.