November 1. Black Friday is 26 days out, the owner wants "our best one ever," the designer works three days a week and the ad budget is the same as October's. If you run marketing alone, or on a team of two, you have seen this movie. AI for marketing does not rewrite the plot. It changes how many hours you spend writing, sorting and building spreadsheets, and gives those hours back to the part that moves revenue: deciding the offer, the budget and what to cut.
This guide follows one full Black Friday campaign, from audience research to the report the Monday after, and shows where AI helps, where it only looks like it helps and where it creates trouble with the FTC, with your email list or with Meta's ad policies. The example from start to finish: Little Bramble, a Shopify store selling kids' clothing out of Columbus, Ohio, with about US$ 60,000 a month in revenue, US$ 8,000 a month in ad spend (US$ 5,500 on Meta, US$ 2,500 on Google) and one marketing person, who is you.
AI for marketers: what it does well and what stays yours
My opinion, after looking inside a lot of ad accounts: AI is a great producer and a poor strategist. It writes 40 ad variations in a minute, summarizes 300 customer reviews, outlines a landing page and turns a table of numbers into a paragraph the owner actually reads. It does not know that the bodysuits carry a 38% margin, that size 2T sold out on Tuesday, or that Meta is finally delivering well on the campaign it suggests pausing.
The split I use is simple. Volume of text, organizing information and first drafts go to AI. Anything that touches money (offer, discount, budget, bids), anything that is a promise to the customer and any decision to pause or scale stays with you. People who flip that split find the mistake on the credit card statement.
The routine, task by task
| Task | Without AI (per campaign) | With AI | Where AI helps | Where it goes wrong |
|---|---|---|---|---|
| Audience research | 4 h | 1 h | Turns reviews, support tickets and ad comments into pains and objections | Makes up market data if you don't give it a source |
| Offer and calendar | 3 h | 1.5 h | Compares offer formats, builds the timeline | Doesn't know your margin or your inventory |
| Ad copy | 3 h | 45 min | Variations by angle, headlines inside character limits | Overpromises, writes lines that break Meta policy |
| Landing page | 5 h | 2 h | Structure, copy, FAQ, a skeptical-customer review | Can't see the page on a phone or measure speed |
| Email and SMS | 3 h | 1 h | Sequences, subject lines, abandoned cart messages | Forgets consent, the unsubscribe and the postal address |
| Content and SEO | 4 h per article | 1.5 h | Brief, outline, first draft | Generic text that reads like every competitor |
| Reporting | 3 h a week | 1 h | Reads the spreadsheet, flags changes, writes the summary | Gets math wrong and trusts the platform's numbers |
Little Bramble's Black Friday, step by step
Audience research: start with what customers already wrote
Asking AI to "describe the persona of a mom who buys kids' clothing" gets you a pretty, useless paragraph. You get the internet average: woman, 25 to 40, cares about quality and price. What works is pasting what your own customers wrote: product reviews, support emails, comments on your ads, DMs. I exported 214 reviews from Little Bramble's Shopify reviews app and asked for three things: the most cited pains, the objections before buying and the exact phrases that could become ad copy.
Something showed up that nobody at the store had noticed: 31 reviews talked about clothes that "survive daycare." Durability, not cuteness. That became the main angle for November. The full prompt, with the real answer, is in ChatGPT prompts for marketing.
One habit: strip names, emails and phone numbers before pasting. The AI doesn't need to know who wrote the review, and if you sell to California residents the CCPA treats that information as personal data even when it sits in a review export.
Offer and calendar: AI runs the numbers, you decide
I gave it the average order value (US$ 58), the average gross margin (52%) and three possible formats: 30% off sitewide, a fixed-price bundle, and tiered discounts (buy 3, save more). It worked out the margin left in each format and named the main risk of each. On a US$ 58 order, 30% off leaves about US$ 12.76 after product cost, roughly 31% of the sale. The bundle of three bodysuits at US$ 39 (US$ 18 each on their own) leaves about US$ 13 per kit, around 33%, and it pushes more units per order. Little Bramble went with the bundle, because it lifts order value without burning margin on single items.
For the calendar, AI lays out the phases well: VIP list, early access, Black Friday (November 27 in 2026), the weekend and Cyber Monday.
And one rule AI rarely brings up on its own: the "was" price has to be a price you actually charged, for a real stretch of time. Raising prices in October so you can show "50% off" in November is exactly what the FTC's Guides Against Deceptive Pricing describe.
Ads: AI for volume, you for the filter
For Meta, I ask for variations by angle, not by wording. Five angles (durability, bundle savings, easy exchanges, gifting, social proof), three short texts each. For Google, I ask for headlines of up to 30 characters and descriptions of up to 90, which are the limits for responsive search ads, and I tell the AI to count. Then I count again.
About the platforms' own automation: Meta Advantage+ and Google Performance Max generate copy and image variations by themselves. I let the platform test audiences and placements, but I review everything it writes in the brand's name. I have seen an automatic variation turn "up to 40% off" into "40% off everything," which was not true.
Reels scripts and organic captions are covered in AI for social media. Here the focus is paid ads.
Landing page: AI builds it, Clarity tells the truth
Little Bramble's Black Friday page was built in a Shopify theme section. AI wrote the structure (promise, proof, bundles, size chart, exchanges, FAQ, deadline) and the copy. Then I asked for something different: read the page as a skeptical mom, phone in one hand and a toddler on her hip, and list everything that would make her close the tab. Three of the points were good: the size chart was buried, the delivery estimate only appeared at checkout, and there was not a single photo of a real kid wearing the clothes.
What AI cannot see: load speed, a button hidden behind the cookie banner, the spot people tap thinking it is a link. Microsoft Clarity (free) records sessions and shows heatmaps. On Little Bramble's page, 22% of mobile taps landed on a bundle photo that didn't link anywhere.
Email and SMS: where AI saves the most and risks the most
The VIP list sequence (5 emails over 10 days) took 40 minutes with AI. I wouldn't send any of it without three checks.
- CAN-SPAM basics. Every marketing email needs a working unsubscribe link, a valid physical postal address and a subject line that isn't misleading. Opt-outs have to be honored within 10 business days.
- SMS consent. Marketing texts sent through Klaviyo, Attentive or Postscript need prior express written consent under the TCPA, and the message should identify the brand and say how to reply STOP. A customer who gave you a phone number at checkout for shipping updates did not agree to promo texts.
- Quiet hours. Don't schedule a "doors open at midnight" text. Keep SMS between 8 a.m. and 9 p.m. in the recipient's time zone, and check your state: Florida and a few others are stricter.
Uploading a customer list to build a Meta Custom Audience or a Google Customer Match list is also a use of personal data. If you sell into California and meet the CPRA thresholds, that can count as "sharing," and your privacy policy and opt-out links need to say so.
Content and SEO: AI writes the average, you add what only the store knows
"How to pick kids' clothing sizes" gets searched all year. A 100% generated article will look like the top ten results on Google, because that is what the model learned from. The article that ranked for Little Bramble had what nobody else had: the store's real size chart, measured on its own garments, the rule from the production team ("if your kid is between sizes, go up one; the cotton shrinks about 3% in the first wash") and photos of the most common exchanges.
Reporting: the best use and the biggest trap
Every Monday, the same ritual: export Meta, export Google, open Shopify and explain to the owner why ROAS dropped. With AI that falls from three hours to one. But there is a trap it won't spot on its own: Meta and Google both claim the same sale. Click an Instagram ad, then search the brand on Google, and both platforms count the order.
I pasted the warm-up week numbers together with Shopify's order count and asked for a short analysis. The answer, unedited:
Week of Oct 26 to Nov 1. The platforms report 152 purchases and US$ 9,896 in revenue on US$ 1,800 of spend (combined ROAS 5.5). Shopify recorded 118 orders and US$ 7,650 in the same period. The gap of 34 orders suggests some sales were credited to both platforms at once. The real return on total spend is about 4.25.
By campaign: Meta retargeting has the lowest cost per purchase outside brand search (US$ 7.50) and a ROAS of 8.6, but it only reaches people who already visited the site; raising its budget a lot will likely exhaust the audience. Google brand search (ROAS 24.4) captures people who were already going to buy and should not be read as a growth channel. Meta prospecting (CPA US$ 17.31, ROAS 3.8) is the one bringing new people in, and it feeds next week's retargeting pool.
Suggestion: keep prospecting spend steady, test a 20% increase on Performance Max (CPA US$ 19.62) and watch retargeting frequency before raising its budget.
Good analysis. I would only hold off on the Performance Max increase until I knew how much of it is branded search in disguise, something the AI had no way of knowing from the data I gave it. I also checked the math in the spreadsheet: all correct this time. Another week, it added up the revenue column wrong. Check every time.
AI tools for marketing and what they cost
| Tool | What I would use it for | Price (Nov 2026, check the site) |
|---|---|---|
| ChatGPT | Copy, spreadsheet analysis, rough creative images | Free with limits; Plus US$ 20/month |
| Claude | Long copy, landing pages, reading big documents without losing structure | Free with limits; Pro US$ 20/month |
| Gemini | If you already live in Google Workspace (Sheets, Docs, Gmail) | Free with limits; Google AI Pro about US$ 20/month |
| Shopify Magic | Product descriptions, email drafts inside Shopify | Included in Shopify plans |
| Meta Advantage+ | Automatic testing of audiences, placements and creative variations | No extra cost, you pay for media |
| Google Performance Max and Google Ads AI features | Campaigns across Google channels, headline suggestions | No extra cost, you pay for media |
| Klaviyo | Email and SMS flows, segments, AI subject lines | Free up to a small list; paid plans by contact count |
| Canva (Magic Studio) | Resizing creatives, background removal, quick text | Free with limits; Pro about US$ 15/month |
| Microsoft Clarity | Session recordings and heatmaps for the landing page | Free |
| Google Analytics 4 + Looker Studio | One dashboard joining Meta, Google and Shopify | Free; third-party connectors charge extra |
| Semrush | Keywords, organic and paid competitors | From about US$ 140/month |
What I would set up with US$ 0, about US$ 20 and about US$ 100 a month
With US$ 0: free ChatGPT or Claude for copy, Clarity for the page, GA4 and Looker Studio for the dashboard, and the AI features already built into Meta, Google and Shopify. You can run a whole Black Friday this way, with daily usage caps in the chat.
With about US$ 20: one subscription, ChatGPT Plus or Claude Pro. I would pick Claude if most of your work is writing (landing pages, email, articles) and ChatGPT if you analyze a lot of spreadsheets and want rough images in the same place.
With about US$ 100: the AI subscription plus a paid Klaviyo plan with SMS. I would put money into email and SMS before a paid SEO tool: for a store the size of Little Bramble, your own list pays back faster than anything else on this table.
Where AI gets marketing wrong
- Market numbers. "The kids' apparel market grows 12% a year" comes out with the same confidence as a real fact. No source, no slide.
- Math. Column sums, weighted averages, ROAS over a period. It gets them right most of the time, and "most" is enough to wreck a report.
- Attribution. It reads platform numbers as truth. Without Shopify's order count next to them, the analysis comes out too optimistic.
- Promises. It loves "the best," "guaranteed" and "instant results." On Meta, copy that asserts a personal attribute ("Struggling with your weight?") gets the ad rejected.
The FTC, CAN-SPAM, the TCPA and platform rules
Marketing written with AI is still advertising, and the legal responsibility sits with the advertiser, not the tool.
FTC Act and pricing. Deceptive ads are illegal whether a person or a model wrote them. On Black Friday the usual problems are invented reference prices, "only 3 left" when there are 300, and countdown timers that reset on every visit.
Endorsements and reviews. The FTC's Endorsement Guides (16 CFR Part 255) require clear disclosure when a creator is paid or gets free product: "#ad" or "Paid partnership," visible without a click. And the FTC rule on consumer reviews and testimonials, finalized in 2024, bans fake reviews and testimonials, explicitly including ones generated by AI. An AI-generated mom saying she "tried it and loved it" is a fake testimonial. Don't make one.
Platforms. Meta requires advertisers to disclose when ads about social issues, elections or politics contain digitally created or altered images or video.
Common mistakes when marketers start using AI
- Asking for strategy without numbers. Without margin, order value, budget and history, AI hands you a textbook plan that fits every company and none.
- Launching 40 ad variations at once. Budget spreads thin, nothing exits the learning phase, and you conclude "AI doesn't work." Launch 3 to 5 per ad set.
- Letting the platform write without review. Auto-generated copy is copy in your brand's name.
- Pasting the whole customer file into a chat. To find buying patterns, remove names, emails and phone numbers first. The analysis comes out the same.
Where to start this week
Do two things with AI before anything else: turn your reviews and support inbox into a list of pains and objections, and build the Monday report with Shopify's numbers next to the platforms'. The first improves every ad you write afterward. The second changes the conversation with the owner, because you stop defending a ROAS that never existed. The 20 prompts used in this campaign, with real outputs, are in ChatGPT prompts for marketing.
Want to skip this part?
Ready-made AI agents for marketers: nothing to build, no prompts to write, they already work for you on every campaign.
- Pulls Meta, Google and your store together every Monday and delivers the weekly report in plain English
- Writes ad variations by angle, inside character limits and checked against claims you can't make
- Drafts email sequences and abandoned cart messages in your brand's voice, with the compliance footer in place
Step by step
- Export reviews and support emails, remove personal data and ask AI for pains, objections and exact phrases
- Give AI your order value, margin and shipping cost and compare offer formats before choosing one
- Write 3 to 5 ad variations per angle and recount every Google Ads character limit
- Have AI review the landing page as a skeptical customer, then watch Clarity recordings
- Draft email and SMS flows and check unsubscribe, postal address and SMS consent
- Build the Monday report with Shopify orders next to Meta and Google numbers
Frequently asked questions
What is the best AI for marketing?
For copy and landing pages, Claude; for spreadsheets and rough images, ChatGPT; for Google Workspace users, Gemini. All three cost about US$ 20 a month on the paid plan, and the free tiers cover a small store.
Will AI replace marketers?
It replaces hours of drafting and spreadsheet work, not the decisions. Offer, budget, attribution and what to promise customers still need someone who knows the business numbers.
Can I use AI-written copy in ads?
Yes, as long as you review it. The FTC holds the advertiser responsible for every claim, and platform auto-generated variations can change what your offer actually says.
Can I use AI to create testimonials or reviews?
No. The FTC rule on consumer reviews and testimonials, finalized in 2024, bans fake reviews and testimonials, including AI-generated ones. Use real customers and disclose paid partnerships.
How much does AI for marketing cost?
You can start at US$ 0 with free chat tiers, Clarity, GA4 and Looker Studio. One paid AI plan is about US$ 20 a month; adding email and SMS software brings a small store to roughly US$ 100.