We showed three AI crawlers the top 7 electricians in New Bedford. The most-reviewed one is invisible — and the cheapest website won.

At 9pm on a Tuesday, when half the house goes dark and the panel smells like warm plastic, a growing number of your neighbours do not type "electrician new bedford" into Google and read ten blue links. They ask ChatGPT. Or they ask Google and get an AI-written answer above the links. Either way, a machine reads the websites and decides what to say about you.

So this month's test is simple: what can that machine actually see of the electricians working New Bedford right now?

First, the respect these businesses have earned, because it's real: the top seven results for "electrician in New Bedford, MA" hold 146 Google reviews between them, averaging 4.91 stars. Not one is below 4.6. You do not get numbers like that from advertising. You get them from showing up, fixing the thing, and charging what you said you'd charge — for years.

That's exactly why what we found matters.

What I actually did#

No opinions in this part. On July 17, 2026, I took the seven businesses Google returns for "electrician in New Bedford, MA", in Google's own order, and did what the AI systems' crawlers do:

• Requested each homepage with the real, published user-agent strings of GPTBot (OpenAI), ClaudeBot (Anthropic) and PerplexityBot — following redirects, exactly as they would.
• Read each site's robots.txt, the file that tells crawlers what they may read.
• Parsed the raw HTML only — no JavaScript executed, because these crawlers don't execute it either. If a fact is painted onto the page by a script, the AI never sees it.
• Checked the raw text for the facts a 9pm caller needs: a phone number, hours, the words "New Bedford", emergency availability, and a Massachusetts license number.

Every check here is reproducible with a laptop and zero budget. The method is the point: this is not what an AI might think of these businesses — that isn't measurable, and I won't pretend it is. This is what an AI can physically read, which is the raw material for everything it says.

Finding 1: The most-reviewed electrician in the set is invisible — and nobody decided that#

Site B is the #2 result. It has the most reviews in the whole sample — 53 of them, at 4.9 stars. It is precisely the business an AI assistant should be recommending.

Here is what a crawler experiences instead. The business's own domain answers with a redirect. Follow it, and you land on a free subdomain of one of the big drag-and-drop website builders. And that subdomain ships a robots.txt that disallows every crawler, of every kind, from the entire site. GPTBot, ClaudeBot and PerplexityBot all publicly state that they obey robots.txt — so they turn around and leave.

It gets worse. Even ignoring robots.txt, the page's raw HTML contains no phone number. The builder paints it in with JavaScript, which no AI crawler runs.

Nobody at this company chose any of that. Someone paid for a website, the builder's free tier made one, and a config file nobody has ever heard of quietly opted them out of the next decade of search. To a well-behaved AI crawler, the best-reviewed electrician near New Bedford does not exist.

Finding 2: Nobody says the things a 9pm caller needs#

For the five sites a crawler can reach, I checked the raw text for the exact facts someone asks an AI assistant during an emergency: who covers New Bedford, are they open, are they licensed, can I call them right now?

Read that middle column pairing again: Site D is a New Bedford company whose website never says "New Bedford" in text a machine can read. An AI deciding whether to mention them to someone in the Whaling City has nothing to go on.

And the license column is the one that should bother everyone. Massachusetts licenses electricians for good reason, every one of these businesses presumably holds one, and it is the single strongest trust signal a homeowner — or an AI ranking answers by trustworthiness — could ask for. Zero of five publish the number. Two say "licensed & insured" as a slogan; a slogan is not a fact a machine can verify.

Finding 3: The cheapest website won, and I'm not sorry#

Here's the opinionated part, since a Reckoning owes you a verdict.

The best machine-readable site in the sample — Site F, rank 6, a small one-man operation — is visibly the cheapest: an off-the-shelf industry template on a shared subdomain. It won anyway, because the template does the boring things: phone, "New Bedford", "emergency", and proper LocalBusiness structured data, all as plain text in the raw HTML.

Meanwhile Site A, the #1 result, serves a grand total of 109 words of machine-readable text. Site E has the most sophisticated structured data of the set — a proper Electrician schema with opening hours — and then never mentions New Bedford. And Site B's polished builder page is, as covered, invisible outright.

So the verdict: the fancy website lost to the boring one, again. If you're choosing between a $3,000 design refresh and an afternoon spent putting plain facts into plain text, the afternoon wins. It isn't close.

One number from the companion speed audit puts a bow on it: of the four sites where Google's PageSpeed test completed, the median mobile Largest Contentful Paint was 16.6 seconds against Google's 2.5-second "good" threshold, and the slowest took 50.8 seconds. The machines aren't the only ones struggling to read these sites. The humans are waiting too.

Mollick's advice is aimed at using AI for your work, and it applies doubly here — because the fastest way to find out what the machines think of your business is to go ask one.

Do this Monday — free, about ten minutes#

First, the sixty-second version of my whole test. Open ChatGPT or any AI assistant and type: "What can you tell me about [your business name] in New Bedford?" Read what comes back. If it's thin, wrong, or confuses you with someone else, you've just seen your website the way the machines see it.

Then the ten-minute fix. Open your website editor — any builder can do this — and put one plain line of ordinary text in the footer of every page:

Licensed MA electrician, Lic. #A12345 · Serving New Bedford, Fairhaven,
Dartmouth & Acushnet · Mon–Fri 7am–5pm · 24/7 emergency · (508) 555-0100

Your real license number, your real towns, your real hours, your real phone number. Text, not an image, not a widget. That single line would have moved every reachable site in this sample to the top of the table. It costs nothing.

If your site came from a website builder, check you're on your own domain and not a free subdomain, then visit yourdomain.com/robots.txt. If you see Disallow: / under User-agent: *, your website is telling every crawler on Earth to leave. That's a setting, and it's fixable today.

What I could not measure#

Honesty section. I did not measure whether ChatGPT, Claude, Perplexity or Google's AI actually recommends any of these businesses — that varies per user and per session and is not measurable from the outside, so I won't claim it. One site's mobile speed test would not complete, so it is reported as not measured, not as a number. And review counts are as Google reported them on July 17, 2026.

I have the names behind every letter in that table. They stay private; none of these businesses asked to be graded in public, and every one of them has earned its stars. If one of them is yours and you'd like your own numbers — the real ones, with the fixes ranked cheapest first — I'll run the full test on your site for free and walk you through it. No pitch attached.

Joshua Amado — Avalon Partner, Fall River 774.559.8992 · Joshua.Amado@AvalonPartner.com

Method: measured July 17, 2026. Sample: the 7 results Google Places returns for "electrician in New Bedford, MA", excluding chains. Crawler checks used the published user-agent strings of GPTBot, ClaudeBot and PerplexityBot, followed redirects, and executed no JavaScript — because the crawlers don't. robots.txt rules were read and reported as written, for every site. Speed figures are mobile Largest Contentful Paint from Google PageSpeed Insights lab data; Google's thresholds, not ours. Anything we could not measure is reported as not measured.