GEO · HOW TO BUY IT WITHOUT GETTING TAKEN
$12,000 a month.Ask them these nine questions first.
That's the public starting price at the top of the generative engine optimization market. Almost every "best GEO agency" list you'll find was published by an agency that put itself on it. Here's how to tell, in one conversation, whether you're buying data or a deck.
WHAT THE MARKET CHARGES
The category priced itself
before it proved itself.
GEO is eighteen months old as a product category. Search demand for the term grew roughly 997% in that time, and pricing arrived well ahead of evidence. None of that means it's a scam — the underlying shift is real and I've written the data up in full. It means you should buy it like you'd buy anything new: by asking harder questions than the seller expects.
Publicly listed starting retainer. Earned media, schema, citation tracking.
Content programme plus technical work, usually on a six-month minimum.
In my own scan of 210 sites, the single biggest differentiator was publishing real prices. That's writing, not retainer.
Ranges reflect publicly advertised pricing in the category as of August 2026. Plenty of these agencies are good at their job — this page is about telling which.
THE INTERROGATION · 1 / 9 OPEN
Nine questions.
Their answers tell you everything.
Ask these on the first call. You're not testing whether they're clever — you're testing whether they've measured anything. Tap any question for the answer you want and the answer that should end the call.
✕ END THE CALL
We'll start seeing improvements in the first month.
✓ KEEP TALKING
Ten fixed prompts, run logged-out across four surfaces, scored and dated before we touch anything.
Why it matters: with no baseline, every future report is unfalsifiable. If they can't tell you today's number, they can't prove tomorrow's.
✕ END THE CALL
We optimise for AI search generally.
✓ KEEP TALKING
ChatGPT, Perplexity, Gemini and Google AI Overviews, reported separately, because they cite different sources.
Why it matters: "AI search" isn't one place. Published citation analyses in 2026 found only around 11% of cited domains overlapped between ChatGPT and Perplexity. Winning one is not winning the category, and a single blended number hides which one you won.
✕ END THE CALL
It's essential — we ship one for every client in week one.
✓ KEEP TALKING
Harmless, no evidence it does anything. We'd rather spend the afternoon on schema and reviews.
Why it matters: this is the cleanest litmus test in the whole category. Ahrefs analysed 137,210 domains and found that of roughly 38,000 with a valid llms.txt, 97% got zero requests for it in a month — no bots, no humans. Google has documented that it has no effect on Search. Anyone still selling it as a deliverable is repeating industry chatter instead of reading data.
✕ END THE CALL
We'll optimise your on-page content and build backlinks.
✓ KEEP TALKING
Most of it. Unlinked mentions, directories, supplier pages, local press, reviews. Your site is the smaller half.
Why it matters: across 75,000 brands, Ahrefs found branded web mentions correlated with AI visibility at 0.66 and backlinks at 0.22 — roughly three times weaker. A vendor still selling link volume as the main lever is running a 2019 playbook with a 2026 name on it.
✕ END THE CALL
You're almost certainly blocking GPTBot — that's the first thing we fix.
✓ KEEP TALKING
Checked it before this call. You're not blocking anything, so that's not your problem.
Why it matters: it's the industry's favourite opening scare. When I scanned 210 real contractor sites, only 5 of them — 2.4% — disallowed any AI crawler. It's catastrophic when it happens and rare enough that leading with it proves they haven't looked at your site. Checking takes thirty seconds; you can do it yourself right now.
✕ END THE CALL
Eight AI-optimised blog posts a month.
✓ KEEP TALKING
Your actual price ranges, your response times, the ten questions you answer on the phone. We'll interview you to get them.
Why it matters: in my scan, the single largest gap between the top-scoring and bottom-scoring sites was publishing real price figures — 87% versus 4%, an 83-point spread. Nobody can write that for you without asking you. If their content plan doesn't start with an interview, it will produce text no model has any reason to quote.
✕ END THE CALL
Results in the first seven days.
✓ KEEP TALKING
Four to eight weeks before anything reaches an answer. Month one is baseline, entity cleanup and the pricing page.
Why it matters: sources have to be re-crawled and re-ingested before a change to your information shows up in an answer. Nobody controls that clock. A seven-day promise isn't optimism, it's a claim about someone else's infrastructure.
✕ END THE CALL
Yes — we have a proprietary method.
✓ KEEP TALKING
No. There's no ranking to buy, no ad slot, and nothing to register with.
Why it matters: ask it deliberately, as a trap. There is no placement to sell and no index to submit to. Anyone who says yes has just told you exactly how much of the rest was true.
✕ END THE CALL
GEO is a long game, you have to be patient.
✓ KEEP TALKING
Here's the number we're moving, the date we judge it, and how you leave if we miss.
Why it matters: "long game" is the most expensive sentence in marketing, because it converts every missed target into a reason to keep paying. A defined metric, a defined date and a defined exit is what separates a partner from a subscription.
WHY I GET TO SAY ANY OF THIS
I scanned 210 real sites
instead of writing a top-10 list.
Every number on this page comes from my own measurement of live U.S. home service websites, or from a study I've linked and named. I run the scanner myself, it's free, and it doesn't ask for your email before it tells you anything.
live contractor sites scored across 17 checks
average quotability — the weakest of four groups
point gap on publishing real prices, top vs bottom
actually block AI crawlers, against the industry scare
The full write-up, with method and caveats, is here. The outside research behind the reasoning is here, with every source linked.
WHAT AN ANSWER ACTUALLY LOOKS LIKE
Twenty contractors in the city.
Three names in the answer.
Ask an assistant who to call for a leaking roof in a named city and you get a short paragraph with two or three businesses in it. Not twenty. Not a page of results to scroll. Three names — and the rest of the city is not in the conversation at all.
So I stopped scoring sites one at a time and started measuring them in groups: one city, one trade, everybody I could reach. Twenty-six groups, 235 companies, the same seventeen checks on every one. The lists are public.
have all three: a price, a response time, an answered question
have none of the three — nothing a model can repeat
answer a single question in plain question-and-answer form
names published per city, because that is what an answer holds
The lists name the top three in each group rather than shaming the bottom — being absent from a short list moves people more than a bad score, and a bad score just gets argued with. If your city is in there, you can see where you land in about eight seconds.
THE WORK ITSELF
Six red rows.
Watch them go green.
This is the actual job, not a service description. Every row below is a check from the scanner, and the number beside it is the gap it closed between the top and bottom of our 210-site scan.
* Starting figure is the median of the weakest third of our 210-site scan. The finish is what the same site scores once these six are done — a measured median, not a best case. The half that depends on reviews and mentions is yours to keep feeding, and no scanner can score it.
THE OFFER
One payment. Six fixes.
A number that moves.
No retainer, no six-month minimum, no monthly report you won't read. The work above is finishable, so it's priced like something that finishes.
AI Visibility Setup
Everything on the list above, done for you, measured before and after. Typically one working week end to end.
WHAT'S INCLUDED
- Baseline scan — all 17 checks, dated, before anything changes
- LocalBusiness schema — hours, service area, phone, services
- Name, address, phone — matched across every profile you control
- A real pricing page — your three most-asked jobs, with ranges
- Response time & hours — in plain text, near the top
- Ten questions answered — Q&A form, FAQPage schema
- Crawler access — robots.txt checked and opened where needed
- Re-scan at day 30 — same checks, same scanner, side by side
Your AI readiness score — same seventeen checks, same scanner, re-run 30 days after we finish — goes up by at least 20 points. If it doesn't, we keep working at no extra charge until it does.
That figure isn't a sales number. Across the 210 sites I scanned, these six fixes moved a weak site by a median of 25 points, and 85% of them cleared 20. I'm promising the number I can back with data, with room to spare.
Who it applies to: sites that score below 70 on the baseline scan — that's where six fixes still have twenty points of room. If your baseline comes back at 70 or higher, I'll say so before you pay anything. A site already in the eighties doesn't need this engagement; it needs one or two specific gaps closed, and that's a smaller job at a smaller price.
What I don't guarantee, and nobody honestly can: that ChatGPT names you next Tuesday. That half runs on reviews and mentions on sites neither of us owns, and it moves on its own clock — four to eight weeks at best. Anyone guaranteeing a spot in an AI answer is selling a placement that does not exist.
WHAT I NEED FROM YOU
- Your real price ranges for three jobs — a fifteen-minute call, and I write the rest
- Access to the profiles you already own: Google Business, Yelp, Angi
- Answers to ten questions you already answer on the phone every week
- All of it inside 14 days — the work is quick, the waiting is what kills it
START
ONE REPLY, FROM ME. NO SEQUENCE, NO NEWSLETTER.
Or skip the nine questions
and just look at your score.
Paste your address. Seventeen checks, ten seconds, no signup, no email wall. You'll know within a minute whether you have a $12,000-a-month problem or an afternoon's worth of writing to do.