Tom McSherry
AI & Search

What AI needs to know about your business before it writes a word

Tom McSherry

Tom McSherry

24 July 2026 · 9 min read

The short version: most bad AI content is not a writing failure, it is a briefing failure. The model can write perfectly well. What it cannot do is know a single thing about your business that you have not told it, and it will never stop and ask. So it falls back on what it does know, which is your trade in general, and the general version is exactly what every competitor in your town is also publishing.

This is for you if you own the business and you are the one pasting prompts into ChatGPT to get your own site finished. Ask for a page on gutter cleaning and you get a competent page on gutter cleaning. What the model does not know is that you will not touch a three-storey job without scaffolding, that half your work comes from property managers, that you charge by the lineal metre, or that what people ring up worried about is whether you will crack the tiles. That is the interesting part, and none of it is in the model.

The model knows everything about plumbing and nothing about your plumbing business. Left to itself it writes the page every plumber could publish, which is the page nobody remembers.

Most owners meet this and reach for a tone instruction. Friendlier, less corporate. Wrong lever, and the reasons are in why ChatGPT content sounds like ChatGPT. The page is not interchangeable because of how it sounds. It is interchangeable because it holds no information that belongs to one business. So the job before you generate anything is getting the particulars out of your head and in front of the model. Here is what to hand over, and how to dig it out.

The particulars only you have

Write these as plain sentences, the way you would explain them to a new staff member on their first Monday. You are briefing, not publishing, so none of it needs polish.

What you actually do, and what you flatly do not

Your real services in the words you use with customers, then the near neighbours you do not offer. The second list matters more, because that is where invention creeps in. A model writing for an electrician will cheerfully mention switchboard upgrades, EV chargers and solar, because electricians do those things. If you do not do solar and the page says you do, you have bought a run of calls you have to turn down. Pull the list off your last twenty invoices, not your existing website: the invoices show what you actually sell, the site usually shows what you meant to sell in 2019.

The suburbs you really cover, not the ones you say you cover

Not the aspirational map with thirty suburbs on it. The places you will genuinely drive to on a Tuesday, the edges where it depends on job size, and the ones you decline outright. Look at where your last thirty jobs physically were. Most owners find their real radius is tighter than their stated one in some directions and wider in others. Write the conditions down too. "Past the harbour bridge only for full installs, not callouts" is a sentence a model can use, and it is what makes a location page read like someone who lives there wrote it.

Your actual prices, and what pushes them up or down

Real numbers, real ranges, or the honest structure if you cannot give a number: callout fee, hourly rate, from-price, minimum job, quoted on site. Then the part almost nobody writes down, which is what moves the price. Access. Age of the building. Two-coat or three-coat. Whether the old unit comes out first. Parking. Left to itself the model produces something soothing about competitive rates and transparent pricing, the most skippable sentence on the internet. A range with its conditions attached lets a reader work out where they sit before they ring. If you would rather not publish prices, say so in the brief and say what to write instead.

The jobs you take, and the ones you turn away

This is my favourite one and it is nearly always missing. What do you say no to, and why? Too small to be worth mobilising the truck. The job type that always ends in a dispute. The cheap imported product you refuse to install because you end up warranting somebody else's decision. Turning work down is the clearest signal that a business has standards, and it filters enquiries before they cost you an afternoon. The model would never guess it, because the general version of every industry says yes to everything.

The objection you answer on every quote

There is a sentence you have said hundreds of times. Why you are dearer than the quote they got last week. Why the cheap fix will not hold. Why the job needs two visits. You have refined that explanation over years of saying it out loud, and it is probably the most persuasive paragraph your business owns, sitting entirely unwritten. Get it down in your own words, including the part where you concede the other side has a point. A page that takes the objection head-on beats one that pretends nobody raises it, because the reader is already thinking it.

What goes wrong on jobs, and how you handle it

Every trade and every clinic has its recurring surprises. The wall opens up and there is no bracing. The part is discontinued. The appointment runs long because the real problem is not the one they booked for. Write down the three or four that come up most, and what you do about each: whether you stop and re-quote, whether you carry the cost, how you tell them. That is what makes copy read as competent rather than merely confident, and it is invisible to a model, which has only read about jobs that went smoothly.

The questions customers ask every week

Scroll the last month of texts and emails on your phone. The same six questions will be sitting there. How soon can you come. Do you charge for the quote. Do I need to be home. Will it be the same person each visit. Those questions with your real answers are the most valuable raw material you have and they cost nothing to collect. They are what customers want to know, which is a different list from what your industry thinks they should want to know.

Your guarantees and terms, in your own words

Exactly what you warrant, for how long, what voids it, deposit and payment terms, cancellation policy, and any licence or registration numbers. Write it the way you say it on the phone, not the way it appears in a supplier's brochure. Keep it accurate above everything else here. A page promising a lifetime guarantee you do not offer is a commitment sitting in public with your name on it, and for anyone registered or regulated this is the section to check hardest.

How you genuinely differ from the business down the road

Be concrete and resist the urge to flatter yourself. "Quality workmanship" is not a difference, it is a claim your competitors make word for word. A difference is structural and checkable: the only one in the district with a particular certification, two people so the owner is on every job, stock on the shelf so nobody waits three weeks, evening appointments. The test is whether the business down the road could copy your sentence onto their site without lying. If they could, it is not a difference, and that is worth finding out before you generate ten pages, because no prompt manufactures one that does not exist.

Who is reading, and what is worrying them

Describe the person who rings you. Homeowner or property manager, first-timer or someone burnt before, urgent or planning ahead. Then the worry underneath the enquiry, because almost nobody buying a service is thinking about features. They are thinking about mess, being upsold, someone not turning up, a bill bigger than the quote. Name the worry and the page can answer it.

Keep it somewhere you will actually reuse it

The mistake after digging all that out is treating it as a one-off, so you are prompting from scratch again three weeks later. Put the lot in one document, call it something obvious like "business brief", and paste it at the top of every session. If your tool supports saved instructions or projects, load it there once and it comes along automatically. What matters is that there is one copy, and you never rebuild it from memory.

  • End it with a hard instruction: if a fact is not in here, ask me instead of filling it in. That changes the default from guessing plausibly to flagging the gap.
  • Keep prices, guarantee terms and coverage in their own sections, so they are easy to correct when they change.
  • Any time you answer the same customer question twice in a week, add it. That is how the brief grows without becoming a project.

How to tell the context is still thin

One test, ten seconds. Read the page you have just generated and mentally put your closest competitor's name at the top. If it still reads as true, the brief was thin. Nothing on that page belongs to you, and a reader comparing three quotes has no reason to pick yours.

The mechanical version is to count the sentences that could only be true of your business. A price and what moves it, a suburb, a job you decline, a thing that goes wrong and what you do about it. On a thin page that count is zero. Two other symptoms of the same cause:

  • It hedges everywhere. Prices "vary depending on your needs", timeframes are "typically quite fast". Hedging is what a writer does when it has no facts.
  • You edit every page heavily. That editing is you supplying the specifics by hand, one page at a time, instead of once at the top.

What this does not fix

Specifics solve the substance problem. They do not decide which pages your site should have or which page owns which search term, and generating at volume without that decision is a separate mess: plan the site first, generate second. They are not voice either, which is covered in giving AI your brand voice. And they do not make the output true, so you still read every number, date and guarantee before it goes live, because a wrong figure reads exactly as smoothly as a right one. The wider structure sits in using AI for your website content without wrecking it.

The honest summary

People ask a stranger to describe their business, then feel let down by the description. The stranger did fine with what it was given, which was nothing but the industry average. So spend the twenty minutes and write the particulars down once. Your prices and what moves them, the jobs you take and the ones you send away, where you really go, the objection you answer on every quote, what goes wrong and how you handle it, the questions that arrive every week, your guarantees in your own words, and the one thing a competitor could not copy without lying. Hand that over every time you generate, and the model stops writing about your trade and starts writing about your business.

Want this done for you?

See how a profit-first SEO strategy could work for your business - no obligation.

See the case studies