Your AI is about to write your Black Friday ads from last spring's prices

Hey Boring Marketer,

Ask your AI what one of your clients is charging for Black Friday.

Does it give you this month's offer? Or the price sheet someone uploaded back in the spring?

On Oct 29, Smári Ásmundsson ran that test live for our community. He used a sample agency with a roofer as a client, and asked its AI what the roofer charges for gutter cleaning and how to say it in an ad.

It came back with a flat $150. The roofer had moved to "starting at $189" weeks earlier, and had asked for "starting at" on every price. The answer named no file and no date, so nobody could check it before it went into an ad.

As Smári put it on his slide: "The model was fine. The sources went bad."

In October that's one wrong ad. In November it's every client at once. Prices, offers and exclusions all change in the same few weeks, the changes land in an email or one person's chat, and then the ads get written fast.

Smári runs his own food brand on more than 150 scheduled AI jobs, and in food a wrong claim on a label is a legal problem. So he built what he calls a company brain: one set of files his AI and his team both read from, where every answer says where it came from.

Below are the 7 parts, in the order I'd set them up before your Black Friday ads go live. Step 6 is the one almost nobody does: telling your AI when the sale ends.

Here's the same question, asked again after Smári's setup ran. This time the AI said "starting at $189," named October's price sheet, and followed the roofer's rule. In Smári's words: "Same model, same question. The only difference is what it was reading."

Most business knowledge gets trapped somewhere only one person can reach. A price sheet uploaded to a Claude project is a copy, so when the price changes, the copy doesn't. Chat memory belongs to one person. A folder on one laptop helps nobody else.

Smári's fix is plain files in storage you already pay for: Google Drive, Dropbox or OneDrive. Prices, client rules, briefs and decisions go there. Orders, customer lists and ad spend stay in their own systems. Then every AI tool your team uses reads from that one folder, and when you switch tools next year you point the new one at the same folder.

Rule: the files are the asset, the app is the reader.

Make one folder per client (or per department if you don't have clients), then split each into an owner folder and a team folder. You share the team folder, never single files, and never through an AI account.

Each person's AI only sees what that person can open. In Smári's example, the part-time contractor gets the roofer's team folder: the brief, the brand standards and the current price sheet. She never sees the agency's margins, the roofer's contract or the other clients.

This matters most in November, when extra writers come in for the rush. Give each one a folder, and when the rush is over, remove one share.

Rule: the folder you're given is the permission.

Every file starts with a short label: what it is, who owns it, when it was last checked, and where the facts came from. The owner is a person, not a team. As Smári said, "Marketing can't answer a question on a Friday afternoon, but Bob from marketing can."

The stamp is what lets your AI end each answer with the file it used and that file's date. So when a Black Friday ad says "starting at $189," you can open the file and see who agreed that price, and when, in about ten seconds.

Rule: an answer that names nothing looks exactly like an invented one.

Anything new goes into one folder: a client's email with their Black Friday offer, call notes with a new rule, an updated price sheet. Overnight, a scheduled AI task files it. It stamps the file, puts it in the right client folder, moves the old version into a versions folder and logs what changed.

In the morning, one person reads a short note, like Smári's: "Replaced the roofer's price sheet. $150 to starting at $189." If the task can't tell which client a file belongs to, it leaves it in the inbox and asks. It never guesses, because a wrong guess changes who can read the file.

Working alone? Skip the schedule. Each morning, tell your AI to "file my inbox."

Rule: nobody should have to remember to file anything.

The roofer's "starting at" rule had been agreed in the account lead's own AI chat, so the writer's AI never saw it. Smári keeps every rule in one rulebook file that every AI reads. Each rule says who set it and when, and can be kept, amended or repealed.

Black Friday exclusions are rules too. "Not on sale: gift cards and new arrivals" has to live where every writer's AI can see it. In June 2025, Australia's consumer watchdog (the ACCC) fined three retailers, Michael Hill among them, over 2024 Black Friday ads that said "sitewide" when some products weren't in the sale.

One warning from Smári: a rule you write at 9 PM when you're frustrated gets followed literally, every day, on every job. Make every rule easy to change.

Rule: if a rule only lives in one chat, the rest of your team's AI can't follow it.

This is my addition, built on two parts of Smári's system.

First, every kind of document gets a review interval: 90 days for a price sheet, a year for house style. Once a file passes its date, the nightly check flags it. A Black Friday offer needs a much shorter clock. File it beside the regular price sheet, not over it, and give it the date the sale ends.

Second, Smári's system never deletes an old document. It marks it as replaced, and the AI stops answering from it. So on the morning after Cyber Monday, the sale sheet gets marked replaced, and your AI goes back to quoting the regular price without anyone rewriting a prompt.

It matters on the way in, too. Under the FTC's pricing guides, a "was" price has to be one you really charged in the recent past. If your AI pulls the "was" price from an old price sheet, the discount it advertises isn't a real one.

Rule: a sale is a rule with an end date.

Before you trust it with a single ad, ask your AI something you know isn't in the files. Smári asked about the roofer's warranty on a new roof. There was no document on it, so the right answer was: "There's no document on the roofer's warranty. The account lead owns this folder. Ask the roofer and drop it in the inbox."

Guarantees, shipping cutoffs and return windows are what an AI invents in a holiday ad. Smári's slide: "An AI that makes up a warranty for a roofer's ad is a real problem. This one won't." If yours answers anyway, fix its instructions before anyone uses it.

Rule: "I don't have that" beats a confident guess.

Once your AI quotes the right price, the ad still needs a page worth clicking through to. Smári's sample client was a roofer, and so is ours: the Keystone Roofing funnel is free (Keystone is a sample roofer we made up). It started from one prompt in BoringFunnels and has the booking page, the quiz, the page that explains the estimate and the follow-ups. You also get a real funnel every week, taken apart from the ad to the last email and ready for your agent to build from. All free with an account, no card.

Inside the Boring Marketing community you'll find:

If you run a food or drink brand, Smári also runs a four-week AI CPG Workshop starting Dec 1.

2,500+ founders and marketers are already building this stuff in there. Come build with us.

Every client's prices are about to change at once. Make sure your AI hears about it before your ads do.

Wanna see a funnel that's selling right now, ad to last email? Get this week's free funnel.

Wanna build alongside 2,500+ marketers? Join The Boring Marketing Community.

Run test one on your own AI and hit reply with what it said. I read every response.

Know someone who'd love this newsletter? Share it with a friend who's building a service for local businesses.

Keep it boring,
James