Most people use ChatGPT to write ad copy, then copy it into Ads Manager by hand. This guide covers something different: building and launching a complete Meta campaign from a single ChatGPT conversation using the approved AdStellar app. ChatGPT on its own cannot see your ad account or launch anything, and plain ChatGPT copy is not tied to your data. The app is what connects the two. Before you start, you need a ChatGPT account that can use apps, an AdStellar account, and a Meta ad account with a connected Page and a payment method. Each step below includes prompts you can copy and adapt.
Step 1: Prepare your Meta ad account and AdStellar workspace
A conversation can only launch what your accounts are ready to run. Check these items in Meta before you write a single prompt:
- An active Meta ad account with no restrictions or unpaid balances.
- A Facebook Page connected to the ad account, plus an Instagram account if you want Instagram placements.
- A working payment method.
- A pixel or dataset installed and firing the right events, if you optimize for purchases.
The pixel check is the one people skip. A conversion campaign with no purchase events has nothing to learn from, and the agent can only optimize against signal that exists. Open Events Manager and confirm your purchase event is recording before moving on.
Next, create or choose an AdStellar workspace for this brand. Because AdStellar supports unlimited workspaces, each store or client can live in its own space with its own data. For agencies, this keeps one client's creatives and performance history from bleeding into another's recommendations.
Then connect Meta inside AdStellar. This gives the agent real-time Meta data and your past performance to work from. If you already run campaigns, import them so the agent's recommendations reflect your actual history rather than starting from zero.
Common mistake: connecting the wrong ad account when you manage several. Confirm the account name and ID in the workspace before you go further.
Step 2: Connect the AdStellar app to ChatGPT
Open the approved AdStellar app listing in ChatGPT: the AdStellar app. Add it, then sign in to authorize access to your workspace. Review the permission screen before you accept, so you know what the app can read and do.
If you use an MCP-capable client instead, the only endpoint to use is https://app.adstellar.ai/api/mcp. MCP (Model Context Protocol) is a standard that lets AI clients call external tools, which is how a chat window can act on your ad account. Do not paste any other URL, and do not trust a third-party connector that claims to be AdStellar. A wrong endpoint could expose your ad account to something you did not intend.
Start a new conversation, make sure the AdStellar app is selected, and run a test prompt:
Using AdStellar, list my connected Meta ad accounts and workspaces.
A healthy response names your workspace or workspaces and the Meta ad account linked to each. If you see your real account names, the connection works.
Troubleshooting:
- If the app does not appear, check that apps are enabled for your ChatGPT plan and workspace. Plan requirements and availability change, so verify the current ChatGPT app requirements as of November 2026.
- Make sure the app is actually selected in the conversation. Having it installed is not enough.
- If the workspace list comes back empty, re-authorize the app and confirm you signed in to the correct AdStellar account.
Step 3: Ask ChatGPT to analyze your past performance and set the goal
Build from evidence, not guesses. Your first real prompt should ask what has already worked:
Analyze the last 30 days of my Meta account in AdStellar. Which creatives, audiences, and placements drove the lowest cost per purchase? Only count ads with at least $200 spend or 10 purchases. What should I scale or cut?
That minimum threshold matters. Judging a winner on a few dollars of spend rewards luck. State a spend or conversion floor in the prompt so the agent ignores ads that have not had a fair test.
The answer draws on AdStellar's AI Insights and element-level creative scoring, which rates individual parts of an ad (the hook, the visual, the headline) instead of only the ad as a whole. The agent also shows its reasoning, so you can sanity-check why it called something a winner. If a conclusion looks off, say so and ask it to re-run the analysis with a different window or threshold.
Before you build anything, decide on one campaign objective and one target. For example: purchases at a $30 target CPA, or a 2.5 ROAS. Put it in writing in the chat:
Our goal for the next campaign is purchases at or below a $30 CPA. Confirm you will optimize toward that.
Vague goals such as "get more sales" produce vague campaigns. A single number gives you something to measure the results against later.
Step 4: Prompt the agent to build the campaign structure
Now ask for the build. Give the agent everything it needs in one prompt: product or URL, objective, budget, geography, audience direction, and what to reuse.
Build a Meta sales campaign for [product URL]. Objective: purchases, target CPA $30. Budget $150/day, US only. Use prospecting audiences based on my best past performers. Reuse top-ranked creatives from my creative library and explain your choices. Build it as a draft. Do not launch.
That last sentence is the most useful habit in this guide. Asking for a draft first lets you edit within the same conversation before any money moves.
When the response comes back, review it the way you would review a media buyer's proposal:
- Structure: the campaign, ad set, and ad hierarchy. Is it what you asked for?
- Budget split: does the daily total match $150, and is it divided sensibly across ad sets?
- Audiences: do they line up with the performers identified in Step 3?
- Reasoning: does each choice come with a stated reason you agree with?
Compare this with manual setup, where you click through campaign, ad set, and ad levels and make each decision from scratch. AdStellar builds a data-backed campaign from your past performance in under 60 seconds. You still approve it, and you should push back where it is wrong:
Move 30% of budget from the broad ad set to the lookalike ad set, and remove the Audience Network placement.
Step 5: Generate creative: images, video, UGC, and clones of winners
If your library lacks enough assets, generate more in the same conversation. Match the tool to the job:
- URL to Ad Maker: turns a product page into ad concepts.
- AI image ads: static creatives with different hooks and layouts.
- Product Video: short videos built from your product assets.
- AI UGC avatars and talking-head ads: presenter-style video without a shoot.
- Ad Clone: reuses the structure of an ad that already works.
Example prompts:
- Create 5 image ads from [product URL], each with a different hook: price, social proof, problem, benefit, and urgency.
- Make a UGC avatar video ad for [product] with a problem-solution script, under 30 seconds.
- Clone the structure of my best-performing ad and swap in our new bundle offer.
Decide what you are testing before you generate. Pick one or two variables, such as hook, format, or offer, so each variation teaches you something. Fifty near-identical ads tell you nothing and split your budget thin.
Before anything goes live, review every asset:
- Brand claims are ones you can support.
- Prices and offers match your site exactly.
- Text is legible on mobile, including in Stories and Reels crops.
- The ad follows Meta's ad policy basics. Policies change, so check the current rules as of November 2026.
Step 6: Expand into a bulk launch of variations
Bulk launch means combining creatives, headlines, and copy into many ad variations at once, then publishing them together. AdStellar can launch hundreds of variations in minutes, which is the part that is painful to do by hand in Ads Manager.
A starting prompt:
Combine these 6 creatives, 4 headlines, and 3 primary texts into variations and add them to the prospecting ad set in this structure. Show me the full list before launching.
Six creatives, four headlines, and three texts multiply to 72 combinations. The arithmetic is why guardrails matter. Cap the total number of variations relative to your budget so each ad can win meaningful delivery. There is no universal ideal number. If $150/day is spread across too many ads, most will never get enough impressions to judge. Ask the agent to help:
Given a $150/day budget and a $30 target CPA, how many variations can this ad set realistically support? Trim the list to that.
Ask for naming conventions too, so analytics stay readable later:
Name every ad as Hook | Format | Offer, for example Social Proof | UGC Video | Bundle.
Mistake to avoid: mixing objectives or audiences in one batch. If one batch includes both cold prospecting and retargeting, or purchase and lead goals, you cannot tell what drove the results. Keep each bulk launch to one objective and one audience.
Step 7: Approve, launch, and monitor from the same chat
Before anything spends money, ask for a full summary and make the agent wait:
Summarize everything you are about to launch: budget, audiences, placements, ads, and schedule. Wait for my confirmation.
Check the summary against what you intended. If it matches, give an explicit instruction:
Confirmed. Launch the campaign as summarized.
Once it is live, keep working in the same conversation:
- Which ads are showing signs of fatigue?
- Pause any ad above $40 CPA after $80 spend.
- Score each creative element and tell me what to make next.
Creative fatigue is the decline in performance that happens when an audience has seen the same ad too many times. AdStellar's creative analytics include fatigue detection, plus goal-based campaign control and 24/7 monitoring, so the agent keeps watching your targets when you are not in the chat. For a second opinion on Meta-reported results, the Cometly attribution integration lets you cross-check against an independent source.
Give delivery two to three days before making major judgments. Meta's learning behavior varies by account, budget, and event volume, so treat that as a rough guide and not a rule. When ads prove themselves, move them into your creative library:
Save the top three ads by CPA to my creative library.
Check the Live Campaign, Then Feed the Winners Back In
Open Meta Ads Manager and confirm the campaign is live and matches what you approved: objective, budget, audiences, placements, and ads. Any mismatch is worth catching on day one, not day five.
After the learning period, return to the chat and run the Step 3 analysis again on the new data. Use the winners from your library as the starting point for the next build, with the same discipline: one goal, a draft first, and a review before launch. Each round starts from stronger assets than the last.
