The panic started on a Tuesday, six days before Cyber Monday
I had six D2C clients, thirty products between them, and a promise I'd made in October that now felt insane: personalized SMS campaigns for every product, every day from Black Friday through Cyber Monday. That's six days, thirty products, and because I'd also promised A/B tests, 180 individual messages. I'm a freelance marketer working out of a co-working space in Austin. I bill $85 an hour. If I wrote these by hand, I'd spend forty hours on copy alone and make nothing after expenses.
So I didn't write them by hand. I used four different AI models—Claude Sonnet 4.5, Gemini Flash, Llama 3.3 70B, and DeepSeek V3—and tracked which one delivered the best open and click rates for different product types and tones. By the time Cyber Monday ended, I had numbers from Shopify and Attentive showing that the model you pick matters more than most SMS guides admit. Not every model writes urgency the same way. Not every audience responds to the same voice.
Here's what worked, what didn't, and how to match the model to the message when you're building Cyber Monday SMS campaigns on a deadline.
The setup: six clients, four models, one framework
Three of my clients sell skincare and wellness (serums, adaptogens, CBD balms). Two sell home goods (candles, linen bedding). One sells minimalist jewelry. Average order values between $45 and $180. All six use Shopify, all six run SMS through Attentive, and all six wanted daily messages starting Black Friday.
I decided to split the work by tone, not by client. Some products need storytelling—why the ingredient matters, what problem it solves. Some need urgency—stock's low, sale ends tonight. Some need brevity because the offer is simple and the audience skims. And some need price-led copy because the discount is the whole point. I assigned one model to each style, wrote five campaigns per product as a test batch, then tracked opens, clicks, and conversions through Cyber Monday.
The models: Claude Sonnet 4.5 for storytelling and emotional hooks. Gemini Flash for urgency and time-limited offers. Llama 3.3 70B for short, punchy messages under 140 characters. DeepSeek V3 for price-led copy with clear discount math. I used Kryotta because I could switch between models without leaving the workspace, and I needed to move fast.
Claude for storytelling: it won on skincare, lost on candles
Claude writes like someone who believes in the product. When I asked it to write an SMS for a $68 hyaluronic acid serum, it opened with "Your skin holds less water every year after 30. This serum holds 1,000 times its weight." That's not hype. It's a fact delivered with enough care that it doesn't feel like a biology lesson. Open rate on that message: 28.4%. Click rate: 9.1%. Conversion rate from click to purchase: 11%, which is high for a serum that's not heavily discounted.
Claude also did well with the CBD balm and the adaptogen powder. Anything where the customer needs a reason beyond "it's cheap right now" benefited from Claude's tone. It sounds human. It doesn't rush. The sentences have rhythm.
But Claude flopped on the candles. I gave it a brief for a $32 pine-and-cedar candle and it came back with "Light this and the room fills with the forest you haven't visited in years." That's pretty. It's also slow. The open rate was fine—24.1%—but the click rate was 4.2%, half what the same candle got with Gemini's "20 hrs left: 30% off pine + cedar, smells like the holidays should." People buying candles on Cyber Monday don't need a memory. They need a deal.
Gemini for urgency: it crushed time-limited offers
Gemini Flash is fast and it writes like it's in a hurry. That works when the clock matters. I used it for every "sale ends tonight" and "only 18 left" message across all six clients. The jewelry client had a pair of gold-fill hoops marked down from $78 to $54 for Cyber Monday only. Gemini's copy: "These hoops sold out last Cyber Mon in 4 hrs. $54 today, $78 tomorrow. Not joking." Open rate: 31.2%. Click rate: 12.7%. Sold 43 pairs by midnight.
Gemini also wrote the best "cart reminder" messages. When someone added linen sheets to their cart and didn't check out, Gemini's follow-up was "You left $120 sheets in your cart. Sale ends in 6 hrs, price goes up $40 after that. Link below." That message pulled a 19.3% conversion rate from people who'd already shown intent. The same reminder written by Claude ("Those sheets are still waiting") converted at 11.8%.
The weakness: Gemini doesn't do nuance. When I asked it to write something for a $150 adaptogen powder, it came back with "Stock's low, price is high tomorrow, buy now." That's true but it sounds like a shout. Open rate was 22.4%, click rate 5.1%. The product needed more explanation than Gemini wanted to give.
Llama for brevity: perfect for low-ticket impulse buys
Llama 3.3 70B writes short. I used it for anything under $40 where the offer was obvious and the customer didn't need convincing. A $28 eucalyptus candle: "30% off eucalyptus today. Ends midnight. [link]". That's 37 characters. Open rate: 26.8%, click rate: 10.4%, conversion rate: 14.2%. The message fit on one line in the SMS preview. People didn't have to open it to know what it was.
Llama also handled the jewelry client's lower-priced studs well. "$18 studs, $12 today, Cyber Mon only. Go: [link]". It sold 67 pairs. The copy is almost rude in how little it explains, but that worked for a product where the photo does the selling and the price is low enough that the risk feels like nothing.
Where Llama failed: anything over $80. I tried it on the linen bedding (queen set, $140 marked down to $98) and the message was "Queen linen sheets $98 today, $140 tmrw. Link below." Open rate was fine—25.1%—but click rate was 6.3% and conversion was 7.1%. People spending $98 on sheets want to know why they should, and Llama doesn't give reasons.
DeepSeek for price-led copy: it did the math so I didn't have to
DeepSeek V3 is weirdly good at discount math. When I asked it to write SMS copy for a skincare bundle (three products, normally $142, Cyber Monday price $89), it came back with "You're saving $53 on this bundle today. That's 37%. Normal price is back Wed." I didn't ask it to calculate the percentage. It just did. Open rate: 27.9%, click rate: 11.2%, conversion rate: 13.4%.
DeepSeek also wrote the best "compare the price" messages. For the jewelry client's $54 hoops, it wrote "These are $54 today. Everywhere else they're $78+. Check." That message got a 10.8% click rate, and I watched the Attentive dashboard show people actually clicking through to other sites to compare before coming back to buy. The skepticism in "Check" worked.
The downside: DeepSeek's tone is flat. It doesn't do warmth. When I used it for the CBD balm, the message was "This balm is $44 today, usually $62. Saves you $18." That's accurate and boring. Open rate was 23.7%, click rate 6.9%. The product needed a bit of story, and DeepSeek doesn't tell stories. It states facts.
What I'd do differently: match the model to the product, not the client
I started this project thinking I'd assign one model per client and keep it consistent. That was wrong. The model should match the product and the message, not the brand. Claude worked for high-ticket skincare and adaptogens. Gemini worked for time-limited offers on anything. Llama worked for cheap, obvious buys. DeepSeek worked when the discount was the headline.
If I ran this again, I'd also test combinations. Use Claude to write the first SMS in a three-part sequence (the one that explains why the product matters), then switch to Gemini for the second message (the one that says the sale ends soon), then Llama for the final nudge (the one that's just a link and a deadline). I didn't have time to build sequences this year, but the data says mixing models would work.
One more thing: I tracked cost per message. Claude Sonnet 4.5 and DeepSeek V3 cost more per generation than Gemini Flash and Llama, but the difference was small enough—fractions of a cent per message—that it didn't matter against the revenue. The jewelry client's Cyber Monday SMS campaigns brought in $14,800 in sales. The skincare clients combined did $31,200. My total AI cost for 180 messages was under $4. The constraint wasn't budget. It was knowing which model to use when.
Questions people ask
Can I use the same model for all my Cyber Monday SMS campaigns?
You can, but you'll leave performance on the table. Claude's storytelling doesn't work for urgency. Gemini's urgency doesn't work for high-ticket products that need explanation. Match the model to the tone the product and offer require.
Do SMS open rates actually vary by AI model, or is it just the offer?
The offer matters more, but the model affects click and conversion rates once someone opens the message. I saw click-rate swings of 4–6 percentage points between models on the same product with the same discount.
How do I know which model to start with?
If the product costs more than $80 or solves a specific problem, start with Claude. If the offer is time-limited or stock-based, start with Gemini. If it's under $40 and the discount is obvious, start with Llama. If the discount percentage is the headline, start with DeepSeek.
Is it worth testing multiple models for every campaign?
Not for every campaign, but for your top five products or your Cyber Monday hero offers, yes. I spent two hours testing five versions of the same message across models and picked the one with the best click rate in the first hour. That's fast enough to be worth it when the revenue is real.
I'll run this test again next year with more clients and longer sequences. For now, the framework works: pick the model that matches the tone your product and offer need, track the numbers, and don't assume one model does everything well. If you want to try the same setup, Kryotta lets you switch between Claude, Gemini, Llama, and DeepSeek without opening four different tabs.
