Plenty of Shopify founders now run a daily ritual: export from Ads Manager, paste into a chat window, ask what to do. It works better than people expect. This page is about the point where it stops working, and it is a specific point, not a vague one.
Direct answer: ChatGPT and its peers are excellent at the thinking and writing parts of paid media: copy, angles, structure, explaining a metric, a second opinion. They fall over on the operating loop, because they cannot see your inventory at variant level, they do not know what happened in the account overnight unless you tell them, and they cannot apply a change with an approval trail and a revert. Operator software exists for that loop. Use the chat tool for the parts it is good at and be honest about the rest.
Where a chat tool is genuinely good
- Copy and angles. Ten headline variants in a house voice, a hook list, a rewrite that drops the discount and leads with the product. Quick and often better than the first human draft.
- Explaining and sanity checking. What is frequency telling me, why did CPM jump, is this campaign structure sensible. A patient tutor that never sighs.
- One-off analysis. Paste a 30-day export and ask which ad sets are carrying the account. It will find the obvious pattern.
- Briefs. Turning a ramble into a designer brief or a media plan outline.
If that is your whole need, stop reading and go use it. Our free Shopify AI prompts work in any chat tool and were written for exactly this.
Where it stops: the five things it cannot see or do
| The job | Chat tool | Operator software |
|---|---|---|
| Know which sizes sold out at 3pm and which campaigns are still paying for them | Only if you paste the inventory and the campaign data and ask the right question, every day | Reads Shopify at variant level and the catalogue identity, and lists the conflicts each morning |
| Hour-of-week evidence for dayparting | Can analyse an export you produce | Ingests the hourly history and keeps it, draws the heatmap, proposes with floors |
| Apply a change | Cannot. You go and do it | Stages the action in a review queue; writes after approval; records intent; reverts |
| Remember last week | Session by session, unless you build a memory system yourself | Holds the store's history, the decisions taken, and what happened after |
| Profit rather than ROAS | If you paste COGS, refunds and fees | Computes P&L from real orders and costs, and says when COGS is missing rather than guessing |
The failure that matters most
A chat tool given a Meta export will confidently tell you the evening is under-performing and suggest dayparting. It cannot know that the evening is weak because the size 12 sold out at lunchtime. The advice is coherent and wrong. This is not a criticism of the model; it did the best it could with what it was shown. It is a criticism of the workflow. A stockout diagnosed as a performance problem is the single most expensive mistake in fashion paid media, and it is one that requires stock and spend in the same view to avoid. That whole trap is in stop-loss vs core-size automation.
"Just build it with ChatGPT and a few scripts"
You can. Some operators do: a script that pulls Shopify inventory, another that pulls Meta hourly data, a prompt that reads both, a rule that writes exclusions. What they end up building is operator software, with the fun parts done first and the boring parts (catalogue identity, revert, approval, learning-phase awareness, multi-store isolation) done never. If you have the engineering appetite, it is a good project. If you have a store to run, it is a distraction. The honest version of this argument, including the ways software oversells itself, is in all-in-one ecommerce AI: real vs decoration.
A chat tool is a brilliant analyst with no hands and no memory of your store. Hire it for that. Do not ask it to be the operator.
How Ralph uses models, for the record
Ralph uses large language models too, for reasoning and drafting. They are the last layer, not the product. The product is the data underneath (variant stock, cover, catalogue identity, campaign history, P&L) and the workflow around it (proposal, approval, write, revert). Which model does what on which plan is on the pricing guide. What it is and is not, in one page: what is Ralph AI.
Questions people actually ask
Can I use ChatGPT to run my Shopify ads?
You can use it to think, write and analyse: ad copy, audience ideas, a second opinion on a campaign structure, a quick read of an exported report. It cannot see your inventory at variant level, it does not know what your campaigns spent overnight, and it cannot apply a change with an approval trail. It is a brilliant analyst with no hands and no memory of your store.
What is the difference between ChatGPT and Ralph?
Context and workflow. Ralph already holds the Shopify catalogue at variant grain, the Meta and Google campaigns, the hourly history and the P&L, and it stages actions into a review queue that can write after approval. ChatGPT knows what you paste into it, this session. One is an operator desk; the other is a very good conversation.
Is Ralph just ChatGPT with my data?
No. Ralph does use large language models for reasoning and drafting, but the product is the data layer and the approval workflow around them: variant stock, cover, catalogue identity, campaign conflicts, exclusions with revert. The model is the last layer, not the product.
When is ChatGPT the better choice?
When the job is a one-off: rewriting a set of headlines, sanity-checking a plan, explaining a metric, drafting a brief for a designer. If you do not run paid media every day, a chat tool and a spreadsheet may be all you need, and it is far cheaper.
Is it safe to paste my ad account data into ChatGPT?
Read the data policy of whatever tier you are on, and never paste customer personal data. Aggregated campaign metrics are usually fine. The bigger risk is not privacy, it is acting on advice from a model that could not see the size run was broken.
If you want stock, ads and the morning list in one place: that is what we built Ralph for. A growth operator for Shopify, with review before anything ships. Not magic. Fewer tabs. See Ralph · Docs