What to Prepare Before You Set Up AI Marketing

Prepare eight things before setting up AI marketing: one goal with a baseline number, a list of where customer data lives and what people consented to, a fact sheet the AI must stick to, ten samples of your voice, admin logins for every account, tracking links, a named approver, and a budget of time as well as money.

The fact sheet is the item most people skip and the one that matters most. AI writes confidently about whatever it's given, and when it isn't given your real prices, seasons and rules it fills the gaps with plausible guesses. Preparation is mostly the work of removing gaps. Allow eight to twelve hours spread over two weeks.

Follow me on Instagram@sagnikteaches

The checklist below is filled in for an illustrative farm shop: a shop with a butchery counter and a small café, about 140 weekly veg box subscribers ordering online, seasonal events like pick-your-own strawberries and pumpkins, and Christmas turkey pre-orders. Swap in your own details as you go.

Connect on LinkedInSagnik Bhattacharya

1. Decide the one job AI marketing should do first

• Pick one goal with one number. Why: "more customers" can't be tested; "40 more veg box subscribers by December" can. How to verify: you can say it in a sentence with a number and a date. Farm shop: grow subscribers from 140 to 180 by 1 December, and cut cancellations from about 9 a month to 6.
• Write down today's numbers. Why: without a before, there's no after. Pull four to eight weeks of data. How to verify: a dated note with the figures and where each came from. Farm shop: 140 subscribers, 9 cancellations a month, 11 new sign-ups a month, email open rate 41%, 2,300 contacts.
• Choose one channel to start. Why: email, social, ads and a chatbot at once means nothing gets done well. How to verify: the other channels are written on a "later" list. Farm shop: email first, because the subscribers are already on it; Instagram second.

2. Find every place customer data lives

• List each source and its size. Why: AI marketing is only as targeted as the data behind it. How to verify: a table with source, number of contacts and last update. Farm shop: online shop (veg box subscribers, 610 past and present customers); email platform (2,300 contacts); café paper sign-up sheets (about 180 names in a folder); the till system (card payments, no emails).
• Record what each group agreed to. Why: people who bought a veg box didn't necessarily agree to event newsletters. How to verify: for each source, a line on how consent was collected. If you can't tell, ask your data-protection adviser before emailing that group. Farm shop: the paper sheets said "keep me posted about events", so those 180 can get event emails, not veg box promotions, until they opt in.
• Remove duplicates and dead addresses. Why: duplicates cause double sends, and many email platforms bill by contact count. How to verify: export, dedupe on email address, re-import, check the count dropped. Cleaning messy data walks through the spreadsheet side.
• Agree the fields you actually need. Why: every extra field is one more thing to keep accurate. How to verify: no more than six to eight fields. Farm shop: email, first name, subscriber yes/no, box size, last order date, interested in events yes/no.

For a fuller method on getting data AI-ready, see how to prepare your business data for AI.

Subscribe on YouTube@codingliquids

The farm shop's finished data map, which took about ninety minutes to put together, looked like this:

The 90 past subscribers who never opted in to newsletters were the surprise. Without the map, the first AI-written "come back to veg boxes" campaign would have gone to them too.

3. Build the fact sheet the AI must stick to

This is a single document of true, current facts that every AI prompt and tool draws from. Without it, AI will invent opening hours, describe a turkey as "free-range organic" because those words usually go together, or promise delivery to places you don't reach.

• Products and prices, with the date they were last checked.
• Opening hours, including public holidays and seasonal changes.
• Delivery rules: days, cut-off times, radius, minimum order.
• Seasonal calendar for the year.
• Allergen and dietary information, or a rule that the AI must direct those questions to staff.
• Claims you can and can't make. Words like "organic" or "free-range" can carry certification or legal meaning; list which you're entitled to use.
• Offers you never run, so the AI doesn't suggest them.

Most owners have this information scattered across notes, emails and their head. AI is good at pulling it together if you give it the mess and ask it not to fill gaps:

Below are my rough notes about my business. Turn them into a fact sheet
with these headings: Products and prices; Opening hours; Delivery;
Seasonal calendar; Allergens and dietary; Claims we can make;
Claims we must never make; Offers we don't run.

Rules:
- Use only what's in my notes. Do not add facts.
- Where something is missing or unclear, write "CHECK:" and a question.
- Keep prices exactly as written.

My notes:
[paste notes]

Before, the farm shop's notes (illustrative):

veg box small 18 large 27, deliver tues + fri, order by sun night. strawbs june-july PYO weekends. turkeys - order by 10 dec, bronze + white, all from the farm. cafe shuts 3 in winter? no deliveries xmas week. we're not organic!! but no sprays on the salad.

After, an illustrative extract of what came back:

Products and prices: Small veg box $18. Large veg box $27. CHECK: are prices per week, and do they include delivery?
Delivery: Tuesdays and Fridays. Order by Sunday night. No deliveries in Christmas week. CHECK: which dates exactly, and what is the delivery radius?
Seasonal calendar: Pick-your-own strawberries at weekends, June to July. Turkey pre-orders close 10 December; bronze and white birds, raised on the farm.
Opening hours: CHECK: does the café close at 3pm in winter, and from which date?
Claims we must never make: Organic.
Claims we can make: CHECK: "No sprays on the salad" — confirm the exact wording you're happy to publish.

The CHECK lines are the valuable part. Answer each one, date the sheet, and store it where every tool and person can reach it. Revisit it monthly and before every season change. For how to use it when reviewing AI drafts, see how to fact-check AI marketing copy before it goes live.

Test the fact sheet before any tool sees it

A fact sheet is only useful if an AI actually sticks to it. Before connecting it to anything, paste it into a chat assistant and ask five questions customers really send, with an instruction to answer only from the sheet:

Answer these customer questions using ONLY the fact sheet below.
If the fact sheet doesn't contain the answer, say "Not in the
fact sheet" and nothing else. Don't guess.

1. Can I get a veg box delivered on a Wednesday?
2. Are your turkeys organic?
3. Is the café open on 26 December?
4. How much is a large veg box with delivery?
5. Do you do gluten-free cakes in the café?

[paste fact sheet]

Illustrative answers from the farm shop's first test: (1) "No. Deliveries are on Tuesdays and Fridays." Correct. (2) "No. We are not organic." Correct, and exactly why that line was worth writing. (3) "Not in the fact sheet." Right answer, and a gap: holiday opening hours were added that afternoon. (4) "A large veg box is $27." Wrong, because it ignored the question about delivery; the sheet didn't say whether delivery was included, so a clearer price line went in. (5) "Not in the fact sheet." Café dietary information was then added as a rule: always refer those questions to staff.

Two gaps and one ambiguity in five questions is typical. Each one would otherwise have surfaced as a wrong answer to a customer.

4. Collect samples of how you sound

• Ten pieces you're proud of: emails, posts, a chalkboard sign, a thank-you note. Why: AI copies examples far better than it follows adjectives like "warm". How to verify: you'd happily send each one again.
• Words and phrases you'd never use. Farm shop: "artisan", "curated", "elevate", exclamation marks in threes.
• Two before-and-after edits. Take an AI draft, rewrite it the way you'd say it, and keep both. This shows the AI the gap it needs to close.

Here's one of the farm shop's before-and-after pairs, which went straight into its voice notes:

AI draft: "Discover the vibrant flavours of the season with our curated veg boxes, lovingly sourced from our fields to your table!"
Owner's version: "First new potatoes of the year went into Friday's boxes. Small ones, still muddy, best just boiled with butter."

Building a brand voice guide that AI can follow turns these samples into a reusable document.

5. Sort out accounts, logins and access

• Make sure every account is in the business's name, with you as admin: email platform, social pages, Google Business Profile, ad accounts, online shop, domain. How to verify: log in to each as admin today.
• Turn on two-step verification everywhere, and keep recovery codes somewhere safe.
• Check your email domain is authenticated. Why: unauthenticated mail is more likely to land in spam. Google's sender guidelines require SPF or DKIM for all senders, and SPF, DKIM and DMARC for anyone sending 5,000 or more messages a day to Gmail addresses. How to verify: your email platform's domain settings should show each record as verified.
• Decide which AI assistant you'll use, and on which plan. Why: customer data should go into business tiers that don't train on your content by default, or into consumer accounts with the model-training switch turned off in privacy settings.
• List the access you'd give a helper, at user level rather than as admin, so it can be removed later.

6. Set up measurement before anything goes live

• Tracking links (UTM tags, labels added to a link so your analytics can tell which email or post a visitor came from) on every link you send. How to verify: click one and check it shows up in your analytics under the right campaign.
• A different discount code per channel for offline-heavy businesses. Farm shop: EMAIL10 in emails, INSTA10 on Instagram, so the till can tell which one worked.
• Conversion tracking on the online shop if you'll run ads. Getting this wrong wastes ad budget; set up conversion tracking before you let AI spend your ad budget covers the setup.
• A holdout group for email automations: 10% of eligible people who don't get the new sequence, so you can see what would have happened anyway.

7. Name who approves what, and how fast

• One named approver per channel, and a backup for holidays.
• A turnaround rule. Farm shop: drafts go in a shared folder on Monday, the owner approves by Tuesday lunch, anything unapproved doesn't go out.
• A "never without a human" list: allergens, health claims, prices, complaints, anything involving a named customer.

8. Budget time as well as money

Free tiers help you rehearse, not run: Klaviyo's free plan, for instance, stops at 250 active profiles and 500 emails a month, so a 2,300-contact list needs a paid plan from day one.

The one-page brief, filled in

Pull the answers into one page. It's what you'd hand to a freelancer, paste into an AI assistant's project instructions, or pin above your desk.

BUSINESS: Farm shop with butchery counter, café and weekly veg boxes.
GOAL: 140 to 180 veg box subscribers by 1 Dec; cancellations 9 to 6 a month.
BASELINE (Aug-Sep): 11 new sign-ups/month, 9 cancellations, 41% opens.
FIRST CHANNEL: Email. Instagram second, from November.
AUDIENCES: Subscribers (140); past subscribers (470); event-only
  contacts from café sheets (180, events emails only).
FACT SHEET: /Marketing/fact-sheet (checked 20 Sep).
VOICE: /Marketing/voice-samples (10 pieces, 2 before/after pairs).
NEVER SAY: organic, artisan, curated; no triple exclamation marks.
NEVER WITHOUT A HUMAN: allergens, prices, complaints, named customers.
APPROVER: Owner (backup: shop manager). Drafts Monday, approval Tuesday 12pm.
MEASURE: UTM links, EMAIL10 code, 10% holdout on the win-back flow.
TIME: 1.5 hours a week. Tools: existing email platform, AI assistant.

The mistake this checklist prevents

Here's how an unprepared launch typically goes wrong, drawn from the kind of errors farm shops make in their first autumn with AI marketing. The owner asked an AI tool to write and schedule a pumpkin-picking email, and it went out to the whole 2,300-contact list on a Thursday evening. Three things then happened. The email said pumpkin picking ran "every day in October"; it ran at weekends only, because the fact sheet didn't exist and the AI filled the gap. It went to the 90 past subscribers who had never agreed to newsletters, two of whom complained. And because the link had no tracking tags, nobody could tell afterwards whether the 60-odd extra visitors that weekend came from the email or from a local social media post the same day.

Each problem maps to one section above: the fact sheet (3), the consent map (2) and tracking (6). None of them was an AI failure. The AI did exactly what it was asked with what it was given.

Starting with no customer list at all

Some businesses reach this point with nothing to automate. An illustrative food truck selling at markets has 2,000 Instagram followers and no email addresses. The checklist still applies, but the order changes:

• Section 2 becomes "start collecting". A QR code on the menu board linking to a one-field sign-up form ("Where we'll be each week"), plus a sign-up link in the Instagram profile. Aim for 200 addresses before automating anything.
• Section 3 shrinks to one page: menu, prices, allergens rule, the week's pitch list, private-hire minimums.
• Section 6 is a single discount code printed on the menu board and used only in emails, so a sale from the list is visible at the hatch.

Two months of collecting at busy markets can build a list worth writing to. Preparation, in that case, mostly means patience.

What you can safely leave until later

Preparation can become its own form of procrastination. Three things feel important but can wait until the first automation is running: a full customer persona document (your subscriber data will tell you more in two months than guesswork can now), a year-long content plan (plan a month at a time), and a comparison of every AI tool on the market (the one bundled with your email platform is usually enough to start). When the eight sections above are done, you're ready to build or to brief someone else.

Further reads

• How to Prepare for an AI Consultation and Leave With a Plan — Turn this preparation into a productive session with an adviser.
• What an AI Consultant Needs From You: Access, Data, and Time — The access, data and time an outside helper will ask for.
• How to Create Customer Personas With AI From Real Data — Build customer profiles from the data you've just gathered.
• How to Plan a Month of Marketing Content With AI in One Hour — Put the fact sheet and voice samples to work straight away.
• Do You Need an AI Consultant to Set Up Your Marketing Automation? — Decide whether to build it yourself once you're prepared.
• How to Measure Whether AI Is Improving Your Marketing Results — Compare results against the baseline you set here.
• Is AI Marketing Automation Worth It for a Small Business? — When AI marketing automation pays off for a small business and when it doesn't yet, with a payback sum, a scorecard and three worked cases.
• Questions to Ask Before You Pay Anyone to Set Up AI Marketing — 30 questions in six groups, with what a good answer sounds like, plus a filled-in comparison of two real-looking quotes for a food truck.
• How Long Does AI Marketing Automation Take to Set Up and Pay Off? — Setup times and payback windows by automation type, a four-number payback formula and a delicatessen's first seven months, month by month.
• DIY or Done-With-You: Who Should Set Up Your AI Marketing? — A scoring table for choosing DIY, done-with-you or done-for-you AI marketing, the setup jobs ranked by difficulty, and what a proper handover includes.
• Best AI Email Marketing Tools for Small Businesses (2026) — Choose between MailerLite, Brevo, Mailchimp and ActiveCampaign with practical tests, cost checks and a campsite example.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Gmail email sender guidelines (Google Workspace Admin Help); Klaviyo pricing page.