What your AI assistant can do in uplads (and what it can't)

Last updated October 2, 2026Auf Deutsch lesen

Once you have connected Claude or ChatGPT, you can hand it a brief the way you would hand one to a colleague:

Launch our summer sale on Meta. The videos are in this Dropbox folder, 50 € a day, Germany and Austria, running until 31 August, linking to /summer-sale.

What comes back is a staged launch in uplads - not live ads.

The normal flow

It reads your account. Ad accounts, existing campaigns and ad sets, your creative library, the pixels on the account, and your Brand Context: positioning, voice, products, audiences and the terms to use and avoid.
It imports the creatives. From a Dropbox or Google Drive link. A folder link brings in every image and video in one go, and files you already have are matched instead of uploaded twice.
It writes the copy. Primary texts, headlines and descriptions in your brand's voice - per creative when the creatives differ enough to deserve it. It also reads your launch defaults and the copy templates your team saved.
It builds the structure. Campaign and ad sets with objective, budget, runtime, the pixel event to optimise for, and targeting: countries, regions, cities with a radius, age, gender, plus real interest and behaviour ids it looks up rather than invents. Everything is created paused and holds no ads, so nothing can spend while you are still reading. It can follow your team's campaign and ad set templates; one saved in an account with another currency comes without its amounts, so they are set fresh for this account.
It stages the launch. You get a review link. Open it and the Ad Launcher is already filled in - media in the Ads panel, ad sets ticked, copy in the Ad Setup card, with a notice above the cards saying the draft is prepared and filled in, and any setup notes it left you behind the info icon. The same draft also waits in the Launch Drafts tab below the launcher.

You launch. Check the screen over, change whatever you want, press Launch. From here it is an ordinary uplads launch.

Before it hands you the link, your assistant can check that the landing page answers and show you Meta's own preview of a staged ad in feed, stories and reels.

What a Meta draft can carry

A draft holds what the launcher sets around the ads, so ask for it in the brief:

The Page and Instagram account the ads speak as.
More text: up to five descriptions, a display link, and Advantage+ creative enhancements.
Per creative or per ad set: its own copy, its own link, URL parameters, display link and button, its own ad sets, and its own placements.
Formats: carousels with 2 to 10 cards, flexible ads, and placement groups.
Catalog ads: a catalog, a product set or a Collection.
Partners and languages: a co-branding partner, and the ads in several languages, translated by your assistant.
Like last time: it reads the landing pages, URL parameters, buttons, Page and ad sets of your recent launches, or stages a past launch again from what it was sent with.

Launch the autumn collection like last time, as a carousel of the six product shots, and add a German version.

Launches on TikTok, Google Ads and Snapchat

The same staging works on the other three networks. The review link opens that network's uplads launcher, filled in, and nothing goes live until you press Launch there.

TikTok. Ads from your library videos, images or image carousels with music, into existing ad groups of one campaign, with identity, captions, landing page, button and ad names. Or a Spark Ad on one of your own posts. A creator's Spark code it never takes: you enter creator codes in the launcher. See Launching ads on TikTok.
Google Ads. A new Performance Max asset group, Demand Gen image and video ads, or YouTube video ads, each with the texts of its channel. Launched paused, they stay paused. It can also start from your existing Meta ads: it copies their images and videos into your library and fits their text to Google's limits. Adding images or videos to running asset groups or to an App campaign is different: Google serves them once it approves them, so that launch is always active. Video files need your YouTube channel connected to the Google Ads account. See Launching on Google Ads.
Snapchat. Each library creative as one Snap Ad in every ad set of one campaign you name, with a shared headline, brand name, call to action and landing page. Images have to be vertical 9:16.

If the campaign or ad sets do not exist yet, your assistant builds them first, paused: campaigns and ad groups on TikTok, campaigns on Google Ads, campaigns and ad sets on Snapchat.

Stage the five new UGC videos into the Spring ad groups on TikTok, with a "Shop now" button to /spring-sale.

Turn our three best Meta ads into a new Performance Max asset group in the Summer campaign.

Changing the draft instead of starting over

If you want something different, say so in the same conversation: more text variants, other creatives, another link, other ad sets. Your assistant changes the staged draft in place, and the draft keeps its review link. It can also discard a draft it staged that you no longer need.

Open the review link again after a change. A tab that was already open keeps the version it loaded. uplads refuses to launch from it and tells you whether the draft was changed, discarded or launched in the meantime.
A draft launches once. Launching the same draft a second time is refused.
Your own drafts stay yours. It only changes drafts an assistant staged and that are still open. A draft you saved in the launcher is left alone.

What it cannot do

It cannot publish ads by default. Two tools publish: launching a draft, and retrying a launch that did not finish. Both refuse unless you allowed it, and launching hands back the review link instead.
It cannot raise spend on its own. Switching something on, raising a budget, loosening a bid or letting a rule do any of that unattended needs the Autonomous launching switch below and your yes in the chat. Pausing and lowering a budget need no switch. What it can change in running campaigns is in this article.
It cannot delete your work. Its cleanup tools only remove what an assistant created in this workspace through uplads, on every network - anything you built yourself they refuse to touch, and you remove those in the platform's own ads manager. On Meta a campaign is deleted when Meta allows it, such as a test campaign that never delivered; one with history can only be archived, and your assistant tells you which.
It cannot exceed your role. Every write runs the same permission and plan checks as the app. A view-only member can stage a draft but not launch it.
It cannot work outside your workspace. No other account's data is reachable.

Everything it does lands in your audit log, marked as coming from MCP.

Letting it launch on its own

If you want a fully remote workflow - brief it from your phone, come back to live ads - you can allow it.

Open Configuration → MCP and scroll to Autonomous launching.
Click Allow launching.
Set a Daily budget ceiling and pick its currency. It is a total for the day: launches, switching things on and budget increases by assistants all count together, and so do the scheduled changes and rules an assistant set up, on the day they fire. The ceiling is in the currency you pick with it: accounts that bill in that currency share one daily total, and in an account that bills in another currency an outside assistant cannot start or raise spend while a ceiling is set. Pausing and lowering stay free, and you can still make the change yourself in uplads. If you set a ceiling before it had a currency, pick one for it now. When the next step would go over it, the assistant stops and hands you the review link instead. Leave it empty for no ceiling.

This is the one setting in uplads that lets something spend money without you clicking Launch, which is why it is off until you turn it on, per workspace, and why every autonomous launch is recorded in the audit log. The same switch lets your assistant switch on paused campaigns, raise budgets, loosen bids and let rules do that unattended, each only after your yes in the chat. You can turn it off again at any time and the assistant goes straight back to staging drafts.

The switch and the ceiling are for assistants connected from outside over MCP, such as Claude or ChatGPT. In the assistant inside uplads, pressing GO on its confirmation card is your yes for everything it does there.

After the launch

uplads keeps its own record of every launch run, and your assistant reads it on every network:

Ad by ad. Which file went into which ad set, the ad id, or the exact error.
Stop a run that is still going. Ads already created stay as they are, so ask it to pause them if spend should stop.
Retry what a failed, partial or stopped run did not create, with the same content. That publishes ads, so it needs the switch and your yes.
Check a Meta launch. It reads every created ad back from Meta and lists what differs from what was sent, and reads the first numbers per ad.
Switch paused Meta ads on later, at a set time, for a sale that starts Monday at 8:00. That is an activation, so it needs the switch and your yes, and uplads checks the switch again when the time comes.

Why did part of yesterday's launch fail? Show me the errors.

Check that the ads of the last launch came out the way they were sent.

Getting good results from a brief

The assistant is only as precise as the brief. What actually changes the output:

Name the numbers. Budget per day or total, start and end dates, the countries or cities. Vague budgets produce vague ad sets.
Give it the creative link once. A Dropbox or Drive folder link is better than naming files - it imports everything and groups the placements itself.
Say which structure to reuse. "Put it in the existing Prospecting campaign" saves it from creating a new one.
Ask for targeting by name. It looks up the real interest, behaviour and location ids and puts them on the ad set. Only what no tool can set ends up in the setup notes on the draft, behind the info icon on the notice in the launcher.
Write the brief in the language the ads should run in, unless you tell it otherwise.

Limits you might hit

The MCP surface has daily budgets on entity writes, drafts and creative imports. They are tripwires against a runaway agent, not a cap on real work - a full brief with a couple of campaigns, a handful of ad sets and fifteen assets uses a fraction of each. If your assistant reports that a daily MCP limit was reached, it will stop and tell you rather than retrying; the limit resets within hours, and you can continue in the uplads UI in the meantime.

Which platforms it covers

All four: Meta (Facebook and Instagram), TikTok, Google Ads and Snapchat. On each, your assistant stages new ads for you to launch and manages what already runs. What it can change in running campaigns on each network is in What your AI assistant can change in running campaigns.