For a few products written one at a time inside Shopify, Shopify Magic is enough: it is free on every plan and sits in the description box. ChatGPT is better once you need one consistent voice across dozens of products, bulk rewrites from a spreadsheet, or descriptions built from photos and customer questions. Many small shops end up using both.
The quickest way to decide is a one-hour test on five of your own products, set out below with sample outputs from each tool. The caveat that applies to both: each writes only from what you give it, and neither checks facts. Shopify's own help pages warn that generated descriptions can include benefits you did not ask for.
The two tools, briefly
Shopify Magic
Shopify's free set of AI features. For descriptions, you open a product in the admin, select the "Generate text" icon in the description field, type a prompt (Shopify recommends at least the product title plus two features or keywords), then edit and save. Shopify's tips page describes tones you can ask for, such as expert, supportive, persuasive, playful and sophisticated. It works in all the languages Shopify supports, and it is desktop only. Sidekick, Shopify's admin assistant, can also write descriptions and edit products, showing you the changes before it applies them, which makes it the in-Shopify option for more than one product at a time.
ChatGPT
A separate assistant you paste information into and copy results out of. The free tier has usage limits; Plus is $20 a month; ChatGPT Business is $25 per user a month billed monthly or $20 billed annually, with a two-seat minimum. What matters for descriptions: Projects that hold your brand-voice notes and instructions, the ability to read photos and spreadsheets, and enough room to work through a whole catalogue in batches.
Decision table: which wins on what
| What you need | Shopify Magic | ChatGPT |
|---|---|---|
| Cost | Free on every plan | Free with limits; $20 a month for Plus |
| One new product, right now | Wins: no copying between apps | Slower: paste in, copy out |
| Rewriting 50+ products | One at a time, or via Sidekick with review | Wins: batches from a CSV export |
| Same voice across the catalogue | Only what you type into each prompt | Wins: saved project instructions and examples |
| Working from photos | Prompt-based; Shopify's help for the generator doesn't describe reading photos | Wins: reads the image |
| Working from customer questions and returns | You summarise them yourself | Wins: can sort months of messages |
| Staff access without extra logins | Wins | Needs accounts, ideally a business plan |
| Writing on a phone | Description generator is desktop only | Works in the mobile app |
| Accuracy | Neither checks facts; both need a human read | |
A fair test on five of your own products
Reviews of AI writers are rarely about your products. An hour of testing tells you more.
- Pick five products of different kinds: a bestseller, a slow seller, something with sizing or compatibility questions, a gift item, and a new arrival.
- Write one fact sheet each: specs, the top buyer question and its answer, what it is not suitable for, and two lines of brand voice. The shared template in writing Shopify descriptions that convert works here.
- Give both tools exactly the same text: paste the fact sheet and the same instructions into the "Generate text" prompt and into ChatGPT.
- Score blind: paste the outputs into a document without labels and have someone else score them.
A filled-in scoring sheet from an illustrative test, 1 to 5 per criterion, for a florist's five products:
| Criterion | Tool A total | Tool B total | Notes |
|---|---|---|---|
| Every fact correct | 21 / 25 | 22 / 25 | Both added one unrequested claim |
| Answers the top buyer question in the first two lines | 16 / 25 | 23 / 25 | A buried it mid-paragraph twice |
| Sounds like our shop | 14 / 25 | 20 / 25 | B had three example descriptions to copy |
| Right length for a phone | 22 / 25 | 19 / 25 | B ran long on the gift set |
| Zero invented claims | 3 of 5 | 4 of 5 | Count, not score |
In that test, tool A was Shopify Magic and tool B was ChatGPT working in a project with the shop's voice notes and three favourite descriptions. The gap on "sounds like us" came almost entirely from those examples, which is worth knowing: give Shopify Magic an example description in the prompt and much of the gap closes.
What each tool gave back for the same product
The product: a bundle of dried lavender and bunny tails, 40 cm, sold as a gift. The fact sheet said the top question was "will it drop everywhere?", with the answer "light shedding in the first week; keep away from radiators". Both outputs are illustrative.
From the "Generate text" prompt with the fact sheet pasted in:
Bring the calm of the countryside indoors with this beautiful
bundle of dried lavender and fluffy bunny tails. Carefully
selected and naturally dried, this long-lasting arrangement is the
perfect gift for any occasion and will add a soothing fragrance to
any room for months to come.What to fix: it never answers the shedding question, "perfect gift for any occasion" is filler, and "soothing fragrance for months" is a claim the fact sheet did not make. The fix is to put the question first in the prompt ("Open by answering: will it drop everywhere?").
Here is that fix applied. The prompt the florist pasted into "Generate text" on the second attempt:
Open by answering: will it drop everywhere? Answer: light shedding
in the first week; keep away from radiators and direct sun.
Product: dried lavender and bunny tails bundle, 40 cm, tied with jute.
Tone: supportive. 60-90 words. No "perfect gift", no scent claims.
Match this example of our style: "Honest about the fuss: eucalyptus
drops a few leaves at first, then settles. 50 cm, sold by the bunch."And what came back (illustrative): "A little shedding in the first week is normal, then it settles; keep it away from radiators and direct sun. This 40 cm bundle of dried lavender and soft bunny tails comes tied with jute, ready to hand over or pop in a jug." The shedding answer now leads, the filler has gone, and the voice is close to ChatGPT's project version. The gap left is small: it dropped "direct sun" once in three runs, so the read-through still matters. That second attempt took about two minutes longer to set up, and the prompt can be saved in a shared note for every dried-flower product after it.
From ChatGPT, in a project holding the shop's voice notes:
Expect a little shedding in the first week, then very little: keep
it away from radiators and direct sun and it will hold its shape
for many months. A 40 cm bundle of dried lavender and soft bunny
tails, tied with jute, ready to give or to drop straight into a
jug. The lavender scent is gentle and fades over time.What to fix: "many months" needs to match what the florist actually tells customers, and "fades over time" is honest but should say roughly how long. Better, but still not publishable without a read.
When Shopify Magic is the right call
- You add a few products a week. The time saved by not switching apps outweighs any quality gap, especially if you paste a standard prompt structure each time.
- Several staff write listings. Nobody needs a new account or a subscription, and the prompt lives in a shared note. The note matters more than the tool. Left to themselves, staff type what's in front of them ("eucalyptus bunch, dried, nice for gifts") and get back three sentences about rustic charm with no length, no care advice and no answer to the question customers ask. With the shared note open beside the product (question first, facts, tone, length, one example description), the same person gets a usable draft on the first try, and listings added on a Tuesday afternoon match the ones written in the big rewrite.
- You need quick drafts in another language, since it works in every language Shopify supports; have a fluent speaker check them.
- You want in-admin bulk help and prefer to stay inside Shopify: that is Sidekick's territory. See using Shopify Sidekick to run your store faster.
When ChatGPT earns its $20 a month
- A catalogue rewrite. Export to CSV, work in batches of ten rows, re-import. This is where ChatGPT saves whole days.
- A distinctive voice. A project with your voice notes, banned phrases and three model descriptions produces more consistent output than a fresh prompt each time. Our guide to a brand voice guide AI can follow shows what to put in it.
- Descriptions from evidence: sorting customer emails and returns into buyer questions, or describing a product from its photo before you write the fact sheet. Photos have a limit, though. Shown a picture of a ribbed green vase, ChatGPT might describe it as "hand-blown glass with a subtle sea-green tint, watertight and ideal for fresh stems". The colour and shape are useful. "Hand-blown" is a guess (it could be pressed glass), and "watertight" is something no photo can show. Use photos for what's visible (colour, shape, texture, proportions) and take materials, capacity and care from the supplier's specification.
- Two or more people writing, where a shared business workspace keeps everyone on the same instructions; the tutorial on ChatGPT Plus vs ChatGPT Business covers when a team plan is worth it.
A realistic mistake: the wrong tone on the wrong product
A florist asked Shopify Magic for a "playful" tone across a batch of new listings to liven up the shop. The batch included two sympathy arrangements, which came back with light-hearted lines about brightening someone's day. They went live for a weekend before a customer pointed it out. The same thing happens in ChatGPT if a project's instructions say "fun and chatty" with no exceptions. The fix is a rule in every prompt or project: "For sympathy, funeral or get-well products, use a calm, plain tone regardless of other instructions", plus keeping those products out of mixed batches.
Where ChatGPT batches go wrong, and the CSV habit that prevents it
Bulk work has one failure the one-at-a-time route never has: descriptions ending up on the wrong products. It happens like this. You paste ten rows, ask for ten descriptions, and the model returns them in a slightly different order, or merges two similar bouquets into one and invents a tenth to make up the number. Copied back into the CSV row by row, three products now carry each other's text. On a florist's shop that means a vase listed with a bouquet's stem count, and it can sit there for weeks.
The habit that stops it: include the Handle column (Shopify's unique product identifier in the export) in every row you paste, tell the model to return a table with Handle as the first column and one row per input row, and match the results back by handle with a lookup formula rather than by position. Then count the rows: ten in, ten out. If the count is off, don't import that batch. For a 90-product catalogue it adds perhaps two minutes a batch.
The hybrid workflow most small shops end up with
- Build the brand-voice project and a standard prompt in ChatGPT, using the five-product test to refine it.
- Use ChatGPT for the one-off catalogue rewrite and for any product with complicated questions.
- Paste the same standard prompt into Shopify Magic for day-to-day additions, so new listings follow the same structure.
- Apply one checking routine to both: facts against the fact sheet, top question answered first, banned phrases absent.
How the choice plays out for a florist with 90 products
Here is how the choice plays out for a hypothetical florist with 90 products on Shopify: 30 bouquet styles, 25 vases, 20 dried-flower bundles, 15 plant-care items. It adds about four products a month, and most existing descriptions are one-line manufacturer text.
- The rewrite: nine batches of ten in ChatGPT Plus, about 25 minutes each including checks, so under four hours. One month of Plus at $20 covers it.
- Ongoing: four new products a month through Shopify Magic with the standard prompt, about ten minutes each. No extra cost.
- Decision: keep ChatGPT Plus for the month of the rewrite and for seasonal ranges (Valentine's Day, Mother's Day), and cancel in between if nobody uses it for anything else.
Over a year the gap is easy to see. Shopify Magic costs nothing beyond the plan the florist already has. ChatGPT Plus kept all year is $240; used only for the rewrite month and two seasonal ranges, it is about $60. Claude Pro on annual billing works out at $17 a month if the owner prefers its writing, but a yearly plan only makes sense if it's used for more than descriptions.
The deciding factor was not writing quality, which the test showed was close once Shopify Magic had an example in its prompt. It was the bulk work: 90 descriptions one at a time inside the admin would have taken far longer than batches from a spreadsheet. A rough sum makes the point. In Shopify Magic, each product means opening it, pasting the fact sheet and prompt, generating, editing and saving: about eight minutes, or 12 hours for 90. In ChatGPT, the same 90 in nine batches came to under four hours including checks. Eight hours saved for $20 is the whole case for paying, and it disappears once the rewrite is done and new products arrive four at a time.
One thing the florist checked before cancelling: whether anything else in the business was using the Plus subscription. The owner had started using it to draft replies to wedding enquiries, which changed the sum. If the subscription is doing a second job, keep it; if the descriptions were its only use, cancel after the seasonal range and restart for the next one.
What neither tool does for you
Neither knows whether the vase is watertight, whether the plant food suits orchids, or whether the product is still in stock. Neither measures whether the new text sells better; you do that in Shopify Analytics by comparing add-to-cart rates before and after. And neither will stop your catalogue drifting into sameness if every prompt is the same, so rotate the banned-phrase list and read five descriptions in a row aloud every few batches.
Shopify Magic and ChatGPT: follow-up questions
Is Shopify Magic really free?
Shopify says its Shopify Magic features, including text generation for product descriptions, are available for free regardless of your subscription plan. You still need a Shopify plan to have a store. There is no separate per-description charge, which makes it the obvious starting point for a shop adding a few products at a time.
Does ChatGPT keep my product data private?
Product descriptions and specs are rarely sensitive, but customer emails and return reasons you paste in to find buyer questions may contain personal details. Remove names and order numbers first. On consumer ChatGPT plans you can switch off use of your chats for model training in privacy settings; ChatGPT Business does not train on business content by default.
Can I use Claude or Gemini instead of ChatGPT?
Yes. Everything in this comparison that applies to ChatGPT, such as bulk batches from a spreadsheet, a saved brand-voice project and working from photos, applies to Claude and Gemini too, and their paid individual plans cost about the same. Pick the one whose writing you prefer after running the same five-product test on it.
Should staff use Shopify Magic even if I use ChatGPT?
Often, yes. Staff adding a single product during a quiet hour do not need a ChatGPT login. Give them the same short prompt structure you use in ChatGPT, pasted into Shopify Magic, so their drafts follow the same pattern, and keep your final check the same for both.
Further reads
- How Much Does It Cost to Add AI to a Shopify Store? — The full cost picture for AI around a Shopify store.
- How to Write Product Descriptions With AI That Actually Sell — Principles of persuasive product copy on any platform.
- Taking a Local Shop Online With AI: Product Listings in a Weekend — Bulk-listing a first range in a weekend.
- Does ChatGPT Train on Client Data? Business vs Free Plans — How training settings differ between free and business plans.
- Tone of Voice Examples: AI Copy Before and After Editing — Before-and-after edits that make AI copy sound human.
- How to Write AI Prompts That Sound Like Your Business — Prompts that keep AI writing in your shop's voice.
- Online Shop AI Mistakes That Hurt Trust and Conversions — Eight AI mistakes that quietly raise returns and lower sales in small online shops, each with a real-looking example and the fix.
- Is AI Frame Advice and Virtual Try-On Worth It for Opticians? — What virtual try-on and AI frame advice cost an optician, the frame-digitising cost most people miss, three practices' verdicts and a 30-day test.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Shopify help pages on Shopify Magic (text generation, product descriptions, tips including tone options) and Sidekick; OpenAI ChatGPT plan pricing as listed in the site fact sheet.