The Machine Recommending Your Competitor

Last month a buddy of mine who owns a commercial roofing company lost a job he never knew existed.

He found out by accident. A property manager he’d done work for years ago mentioned, over a beer, that her new building had gone with somebody else. He asked why she didn’t call him. She said she didn’t think of him. She’d asked ChatGPT for the three best commercial roofers in the county, called all three, and hired one of them.

He wasn’t on the list.

Twenty-two years in business. Four hundred jobs. A shelf full of plaques from the builders association. And a piece of software decided he didn’t exist.

That’s the story I want you sitting with this Monday, because it’s happening to you too. You just haven’t had the beer yet.

For about twenty years the game was simple. Somebody needed what you sell, they typed it into Google, they clicked a few results, they read some reviews, they called. You could fight for that page. You knew the rules.

That front door is moving, and it’s moving fast.

BrightLocal’s 2026 consumer review survey found that 45 percent of consumers now use ChatGPT or another AI tool to get local business recommendations. Last year that number was 6 percent. That’s not a typo. It went from a rounding error to almost half the market in twelve months.

Yext’s 2026 search behavior report found something that should matter even more to the guys reading this. Among North American consumers earning $150,000 or more, more than half now start their local search with AI, ahead of Google. Those are your best buyers. The ones with budgets. The ones who don’t haggle over every line.

And here’s the part nobody talks about. When somebody asks Google, they get ten blue links and they make the call. When somebody asks AI, they get a short list. Three names, maybe five, with a tidy paragraph on why each one is good. The machine already did the narrowing for them.

If you’re on that list, you’re in the final round before you ever knew there was a game. If you’re not, you never got invited.

You don’t need a computer science degree to understand this. Strip it down and these tools are doing roughly what a sharp assistant would do if you told him to find you a roofer by tomorrow morning.

They read what’s out there about you, they look for consistency, and they look for proof.

Specifically, they lean on a few things:

Notice what isn’t on that list. How long you’ve been in business, unless it’s written down somewhere. How good your work is, unless somebody said so in public. How many people love you, unless they said it where a machine can read it.

That’s the trap my roofer friend fell into. Everything great about his company lived in people’s heads. The machine can’t read heads.

Before you fix anything, find out what’s being said. This takes about forty-five minutes and it’s going to sting a little.

Open the major AI tools. ChatGPT, Claude, Gemini, Grok, Perplexity. I run them side by side in Galaxy.ai because it’s one login and I can paste the same prompt into every model at once and compare the answers. Use whatever you like, but test at least three, because they don’t all say the same thing about you.

Then ask the questions your buyers would ask. Not your company name. Their problem. Here’s the set I use:

Write every answer down in a simple sheet. Four columns: the question, the tool, whether you showed up, and who showed up instead.

Then look hard at questions four through six, the ones that name you. This is where you find out what the machine thinks you are. Sometimes it’s flat wrong. It’ll say you do residential work when you quit residential in 2019. It’ll dig up a two star review from 2017 and hand it to every prospect who asks. It’ll describe you in a sentence so bland your own mother wouldn’t hire you.

Don’t argue with it. Just write it down. That’s your starting line.

Now take the competitors who showed up instead of you. Pick the two that appear most often. Then ask the AI the dumbest, most useful question there is:

“Why did you recommend [competitor]? What sources are you basing that on?”

Some tools will show you citations. Some will just tell you. Either way, you’re about to get a list of places your competitor exists that you don’t. A directory you never claimed. A business journal article from last spring. A “best of” list from the local chamber. Seventy-three reviews from the last six months against your eleven.

This is the cheapest competitive research you’ll ever do. Your competitor didn’t hand you his playbook. The machine did.

Most of this is unglamorous. Good. Unglamorous is where nobody else bothers to show up.

Get your basic facts identical everywhere. Business name, address, phone, hours, service area, and a one sentence description. Same words, every site. Google Business Profile first, then Apple Business Connect, Bing Places, Yelp, Facebook, and the two or three directories that actually matter in your industry. BrightLocal found Apple Maps jumped from 14 percent to 27 percent of review use in a single year. If you’ve never touched your Apple listing, you’re missing a quarter of the room.

Rewrite your homepage’s first paragraph. Read it out loud like you’re a stranger. Does it say what you do, who for, and where, in plain English? If it says “We deliver innovative solutions that empower our partners,” burn it down. Try this format instead:

“[Company] is a [what you are] serving [who] in [where]. We specialize in [the two or three jobs you most want to be hired for]. Since [year], we’ve [one concrete proof point with a number in it].”

Boring? Absolutely. It’s also exactly the kind of sentence a machine can lift and repeat to somebody who’s ready to buy.

Build one page per money service. If you want more of a specific job, it needs its own page on your site. Not a bullet on a services page. A real page that covers what the job is, who it’s for, how you do it, roughly what it costs, and a short case study with real numbers. When somebody asks AI about that exact job, you want a page that answers that exact question.

Add an honest FAQ. Take the questions your team hears every week and answer them straight. How much does it cost? How long does it take? What happens if something goes wrong? Do you work with companies our size? These tools love a clean question with a direct answer, because that’s exactly what they’re trying to hand people.

Reviews are the biggest lever you’ve got, and most owners run them like a suggestion box by the door.

The BrightLocal numbers are brutal. 74 percent of consumers only care about reviews from the last three months. 47 percent won’t use a business with fewer than twenty reviews. And 31 percent will only use a business rated 4.5 stars or higher, up from 17 percent last year. The bar is climbing, and your glowing review from 2021 isn’t helping you clear it.

So make reviews a system, not a mood.

Ask at the moment of peak relief. Not when you send the invoice. When the client says thank you. The job’s done, the problem’s gone, they’re happy. That’s your window, and it closes fast.

Make it stupid easy. Text a direct link to your Google review page. One tap. No hunting around.

Ask for specifics. This is the move almost nobody makes, and it’s the one that matters most right now. Don’t say “Would you mind leaving us a review?” Say this:

“Would you mind leaving us a quick Google review? If you can, mention what we did for you and how it turned out. It helps the next person in your situation find us.”

That one sentence changes the review you get. Instead of “Great guys, highly recommend,” you get “They rebuilt the HVAC in our 1960s office building without shutting down a single floor.” The first one is a compliment. The second one is a fact. Machines repeat facts.

Respond to every single review. BrightLocal found 89 percent of people expect owners to respond, and half are turned off by canned replies. So write a real line. Name the job. Thank them for something specific. It takes thirty seconds and it shows every future reader that a human is home.

Your own website only gets you so far. What really moves the needle is other people saying good things about you in public, in places that get read.

Pick two of these and work them every month:

Put a recurring thirty minute block on your calendar for the first Monday of every month. Re-run the same six questions in the same tools. Update the sheet. Track three things: how many questions you show up for, where you land in the list, and whether the description of you is accurate.

This is slow. You won’t see the needle move in a week. Think of it like interest. Every honest review, every consistent listing, every article that quotes you is a deposit. Six months from now you’ll run the test and see your name in answers you weren’t in before. And once you’re in, you tend to stay in, because the machine keeps finding more proof that you belong there.

“My clients come from referrals, not search.”

So did my roofer friend’s. Then his best referral source asked a robot instead of her memory. Even referred buyers check you out before they call. Yext found 53 percent of people go search Google or Bing after an AI recommendation, and 49 percent go straight to the company’s website. Your referral is getting fact checked by a machine whether you like it or not.

“This sounds like SEO all over again.”

Some of it is. The difference is the stakes. Page two of Google was bad. Not making a three name shortlist is invisible. And the fundamentals here are things you should’ve done anyway: a clear website, honest reviews, consistent facts, and being known for something specific.

“I don’t have time for this.”

The audit is forty-five minutes. The foundation is one Saturday morning. The upkeep is thirty minutes a month plus asking happy clients for reviews, which you should already be doing. Stack that up against the job you lost and never heard about.

Three things before Friday:

Wednesday we’re leaving the screen behind. In three weeks the clocks fall back, the dark shows up before dinner, and most guys lose the next four months without even noticing. Not this year.

Refined. Relentless. Unapologetic.

Marcus