AI for Meta Ads: What Agencies Hand to Advantage+ and What They Keep

Hand Advantage+ the repetitive optimisation: delivery, audience expansion, placements and creative testing across many assets. Keep the decisions that carry client risk: budget limits, hard exclusions, brand-sensitive creative, offers and claims, and the measurement baseline. Meta switches leads campaigns to Advantage+ by default and steers Sales and App the same way, so the real task is deciding what to override.

Only Meta's own product pages rank well for this topic, and they describe what the automation does, not what an agency should retain. That leaves a gap you can fill for your clients: a written rule for each decision area, a reason, and a way to check the system is behaving. Below are that checklist, sample client wording, a worked example with numbers, and a fix for the attribution change Meta announced on 3 March 2026.

Follow me on Instagram@sagnikteaches

What Meta says Advantage+ automates, and what it does not mention

Meta's Advantage+ page describes end-to-end solutions that apply AI to audience, placements and budget, and lists three campaign types: Advantage+ sales campaigns (formerly Advantage+ shopping campaigns), Advantage+ app campaigns and Advantage+ leads campaigns. Single-step Advantage+ options are available across objectives. Meta's leads page adds that every leads campaign is Advantage+ On by default, with Advantage+ audience, placements and campaign budget enabled, and that narrowing settings too much switches it Off. Meta's developer documentation ties the Advantage+ state to the same three levers for Sales and App campaigns and notes that placement exclusions set at ad-account level leave a campaign eligible, which is the practical detail: put brand-safety exclusions at account level, not on the ad set. None of these pages says which controls an advertiser should keep. That silence is why agencies need their own rule set: if you do not decide what you hold, the default decides for you.

Connect on LinkedInSagnik Bhattacharya

A separate creative page lists features such as image expansion, background generation for catalogue ads, video expansion for Facebook Reels, related media, image animation, text generation, image generation and image touch-up. Meta states that some enhancements may be on by default, that any can be turned off at any time, that you should preview samples before publishing and that previews show examples rather than every variation. Features that use AI carry an AI logo beneath their name, and an information icon explains what each does. Menus and names move, so treat this list as a starting point and check Ads Manager on the day you build.

Subscribe on YouTube@sagnikteaches

The hand-over versus control checklist

This is the core of the rule set. For each decision area, the table states the default I would use, the reason, and what to check. It is a starting position for a typical small-business client, not a universal answer. Adjust where the account's data or the client's risk says otherwise.

The last row is a rule, not a recommendation: do not hand compliance to automation. Meta runs its own review, but the advertiser carries the responsibility. Special categories and sensitive sectors have additional rules; where a client works in one (housing, health and wellness, financial offers), read Meta's current policy page for it before launch and again when enhancements or generated text are switched on.

Delivery and audience: let it expand, but fence it in

Audience is where handing over costs least and gains most, because expansion finds people your targeting would not have named. Meta's system treats your audience suggestions as signals to start from and expands beyond them when it predicts better results. What it cannot be allowed to do is cross the lines the client cares about: location, minimum age, language, and exclusions such as existing customers or staff.

A practical rule: write the fences into the build document, and check them in the first 48 hours, not at the monthly review. Open the delivery breakdown and confirm that spend is not appearing in locations or age bands outside the brief. If it is, the setting was either not applied or is a suggestion rather than a control, and you need to find out which before more budget goes.

Placements and the inventory filter

Meta's inventory filter has three levels, Expanded (the default), Moderate and Limited, which decide how close to sensitive content your ads may run; Limited also excludes live videos, and the filter covers feeds, Reels and Audience Network rather than every placement. Account-level controls such as block lists apply even when Advantage+ placements are on. For a brand-safe client such as a school or a charity, set the filter to Limited and record it. For a removals firm that only wants leads, Expanded is usually fine. The point is that you chose it, and it is written down, so the client never learns about a placement from a screenshot sent by a friend.

Creative: what to feed it, what to switch off

Advantage+ is only as good as the assets you provide, and it multiplies whatever you give it. If you upload six near-identical images, it will find little to learn. If you upload four different messages in three formats, it has something to test. A starting set for one ad set: four to six distinct concepts (different message angle each, not different colour backgrounds), each in square and vertical formats, with three headline options and two descriptions. That is a test design, not a guarantee, and the right number depends on the client's budget.

Then decide each creative enhancement on purpose. Here is an illustrative decision for four clients.

Run the previews Meta provides before publishing and screenshot them for the client file. Previews are examples of possible optimised versions, not a complete list of what may run, so also check live ads in the first days. If you use ChatGPT, Claude or another tool to write your own variants before upload, run the policy pass described in Facebook ad variants written with ChatGPT and checked for policy. Meta's labelling of AI-generated or AI-edited content has changed more than once, so check its current help page on labels, and tell clients what applies to their ads.

Budget and bidding: two rules that catch people out

Budget is the area where I would keep the most control, because it is the client's money. Advantage+ campaign budget spreads spend across ad sets for you, which is useful, and a day's spend may run over the daily figure. Meta's help page on daily budgets says it may spend up to 75 per cent over the daily budget on a given day, while keeping each Sunday-to-Saturday week within seven times the daily budget. Tell clients that a $50 daily budget can show $87.50 on a busy day, though the week stays within $350, and agree a monthly ceiling you can monitor.

Bidding strategy carries its own rules. Meta's help pages on cost per result goal say the strategy works best when an ad set gets at least 50 to 100 conversions a week, suggest starting the goal 10 to 20 per cent above your recent average cost per result, and warn that it may spend less than the full budget to protect the target. A quick sum shows the trap for small accounts: a $50 daily budget with a $20 target cost per lead can buy at most about 17 leads a week ($350 divided by $20 is 17.5), far short of 50, so the system has little to learn from and delivery can stall. Meta's own suggestions are to broaden the audience, raise the budget or raise the cost goal, and with no history it suggests running the Highest volume strategy for at least two weeks to set a baseline. Blaming the creative is not on the list. Wait at least seven days after changing a goal before judging it, and write the strategy and target into the build document.

For a cautionary view on judging a campaign too early, how much to spend on AI-run ads before judging results sets out a sensible learning window.

Measurement: the March 2026 change and your baseline

On 3 March 2026 Meta announced that click-through attribution for website and in-store conversions would count link clicks only. Conversions that followed shares, saves and other non-link clicks, which used to count as click-through, move to what Meta first called engaged-view attribution and now calls engage-through attribution. Meta said the change would begin later that month for campaigns optimising for website or in-store conversions and that advertisers might see it take effect at different points, so check the attribution settings at ad-set level and the comparison view in Ads Manager to find when it reached each account. Meta's stated aim is closer agreement with third-party analytics tools; the practical side effect is that click-through figures can fall when fewer interactions qualify, with no change in real sales.

This matters for an agency because a month-on-month comparison across the change shows a fall that is purely definitional. Handle it in three steps. First, set a new baseline from the first full month after the change, and say so in the client report. Second, reconcile Meta's numbers to the client's own records every month. Third, report the ratio between them, because a stable ratio means the platform is behaving consistently even if the absolute figure changed.

An illustrative case. A removals firm's report in February showed 60 website enquiries at $22 each on $1,320 spend. In April the same campaign shows 41 enquiries at $32, and the client panics. The firm's own booking sheet shows 38 booked jobs in April against 36 in February. Nothing worsened: 36 bookings from $1,320 is $36.67 per booking, and 38 from, say, $1,300 is $34.21. The platform number moved because of a definition change, and the client's sheet shows the truth. Reporting both, with a one-line explanation, turns a scare into a credible agency habit.

A worked example: a florist, a $40 daily budget and a sales campaign

Say a florist wants more online orders for a seasonal range and gives you $40 a day for a Sales campaign, with a monthly ceiling of $1,300. You choose Advantage+ sales because the objective defaults to it and the catalogue is already connected. Here is the hand-over and control split, row by row.

1. Handed over: audience expansion from the customer list as a suggestion, Advantage+ placements, and delivery optimisation to purchases.
2. Kept: location limited to the delivery area, a block list of competitor pages loaded at account level, inventory filter at Moderate plus the News topic exclusion that Meta offers for Facebook in-stream reels and ads on Reels (the client dislikes news adjacency), and exclusion of past buyers in the last 30 days.
3. Creative: five concepts supplied (the product, the delivery promise, a gift occasion, a price-led offer, a care-instructions angle), each approved by the client in writing. Image expansion on; text generation off, because delivery cut-off times must be exact.
4. Budget: campaign budget of $40 a day, monthly ceiling of $1,300 tracked weekly. A daily overspend is explained to the client in advance.
5. Measurement: weekly check of Meta purchases against the shop's order export, with the ratio recorded.

After two weeks, Meta shows 31 purchases and the shop shows 27 orders from paid social, so the ratio is 0.87. In week four, a new batch of images raises Meta's figure to 40 and the shop to 34, ratio 0.85. The ratio is stable, so the agency reports confidently. If the ratio had dropped to 0.5 with no change in the shop's orders, the agency would have looked at duplicated events or a tracking fault before touching the creative. Time cost for the agency: about 45 minutes per week for the checks, which is the price of keeping control.

Leads campaigns: cheap leads that never answer the phone

Leads campaigns carry a failure the Sales examples do not: the system optimises for the form submission it can see, and a form submission is not a customer. Left alone, an Advantage+ leads campaign tends to find the people who submit forms easily, which can mean low intent. A photography studio taking enquiries for family shoots might see 48 leads in a fortnight at $9 each and then find that only 6 replied to the follow-up. Cost per reply is $72, not $9, and that is the number that matters.

The agency's controls here are about quality, not volume. Add a qualifying question to the instant form (a preferred month, or a rough budget band) so that submitting takes a little effort. Track the follow-up outcome, even by hand in a shared sheet: replied, booked, not interested. Report cost per reply and cost per booking next to cost per lead. If your client's CRM can send a later event such as "qualified" back to Meta, ask whether it is worth optimising towards that event once you have enough of them to learn from; before that, a low-volume event will starve the system. Advantage+ leads for a local service business walks through instant forms and budget for this type of build.

Write the rule into the hand-over table for leads accounts: handed over is delivery and audience expansion, kept is the definition of a good lead and the follow-up record that proves it. The same logic applies to a removals firm, where a lead without a move date is a name on a list. Add "move date" as a required field, and the lead count may fall while booked jobs stay level or rise.

Where agencies lose control without noticing

• Settings that reappear. Enhancements or expanded settings can come back on when you duplicate a campaign or when Meta changes defaults. Add a pre-launch toggle check to every build and every duplicate.
• Different people, different habits. Two account managers handle the same client differently, and no one writes it down. A one-page rule set per client fixes that, and it doubles as an onboarding document.
• Creative fatigue hidden by averages. Advantage+ can keep one winning asset carrying a campaign while others get almost no spend. Review spend per asset, not only totals.
• Client expectation drift. A client sees an ad variant they did not approve (a cropped image, a generated line) and loses trust. The approval log with preview screenshots is your defence.

For the plain-English mechanics behind these behaviours, how Meta and Google use AI to run your ads is the background read, and Meta's small business AI features and which to switch on covers the wider feature set.

Client wording that sets expectations

A short paragraph in the proposal or monthly report does a lot of work. Adapt this:

How we use Meta's AI. We let Meta's Advantage+ tools handle delivery, audience expansion and
placement choice, because they usually find buyers faster than manual targeting. We keep
control of your budget ceiling, location, exclusions, the wording of every claim and offer,
and which creative enhancements are switched on. Every image and line is approved by you
before launch. Meta reports its own results; we reconcile them to your sales records each
month, and since Meta's March 2026 attribution change we compare against a new baseline.

Clients rarely object to this. They object to surprises. If you want to see how the same expectations are set in a broader agency rollout, an AI rollout plan for a ten-person marketing agency covers training and review habits across a team.

A monthly review you can run in an hour

1. Pull Meta's reported results and the client's own sales or lead records for the same dates.
2. Calculate the ratio and compare with last month's.
3. Open the placement and location breakdown; compare with the brief's fences.
4. List spend per creative asset; retire the bottom quarter, add two new concepts.
5. Check enhancement toggles on live ads against your approval log.
6. Confirm total spend against the monthly ceiling and note any overspend days.
7. Write three sentences for the client: what changed, what we learned, what we will test next.

Keep the notes with the build document. After three months the file shows which hand-overs paid off and which you reversed, and that is the evidence you need for the next conversation about spend. If you would rather map the rule set to your own client mix with someone, that is what my AI implementation consultation covers, through the contact page. The AI guide lists the rest of the series.

Questions agencies ask about Advantage+

Can we still run manual campaigns for Sales, App or Leads?

Yes. Meta switches leads campaigns to Advantage+ On by default and builds Sales and App campaigns through the same streamlined set-up, but narrowing the audience or excluding placements on the ad set turns Advantage+ off. Look at what Ads Manager offers on the day you build, since menus change, and record any setting you override and why.

Should creative enhancements be switched off by default?

Meta says some enhancements may be on by default and any can be turned off. For brand-sensitive clients, switch them all off and test one at a time with the client's approval, using Meta's previews. For low-risk accounts, a controlled test usually beats a blanket rule.

Why did our reported click-through conversions drop in March 2026?

Meta announced on 3 March 2026 that click-through attribution for website and in-store conversions would count link clicks only; conversions after shares, saves and other non-link clicks are reported as engage-through. The change reached accounts at different points, so find your date, set a new baseline from it and compare against your client's own records.

How much creative should we supply to Advantage+?

Supply enough distinct angles for the system to learn from: several different messages and formats, not several near-identical images. Four to six genuinely different concepts per ad set is a reasonable starting test, then judge on the client's own sales or leads data, not on Meta's dashboard alone.

Further reads

• AI Social Media Monitoring: A Monthly Insight Report for Clients — Another monthly client report that needs honest framing.
• Can AI Manage Your Google Ads Without an Agency? — The same hand-over question on Google's side.
• How to Label AI-Generated Images and Video on Social Media — Labelling AI-generated images and video on social platforms.
• Pressure Washing Facebook Ads: Before-and-After Creative Made With AI — A starting budget with the sums behind it, three offer angles, honest before-and-after photos, and the Meta settings that can quietly alter them.
• Ads in AI Overviews: A Pre-Spend Checklist for Ad Agencies and Clients — How Google ads in AI Overviews are bought and reported, plus a client pre-spend checklist, eligibility checks and a sensible test budget.
• Using AI in Advertising: What a Small Agency Can Hand to a Tool — A task-by-task map of what a small ad agency can hand to AI, where each approval gate sits, and the jobs that stay human.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Meta's Advantage+ overview, creative and leads pages; Meta's 3 March 2026 announcement on click attribution; Meta's business help articles on daily budgets, cost per result goal and inventory filter; Meta's developer documentation on the Advantage+ campaign experience; all read November 2026.