AI uses for Meta ads, ranked
The ranked list
25 AI uses in four groups: (a) Meta's own AI, (b) agents on the ad account, (c) third-party tools, (d) the work around the ads. Each group has rows.
Rule: score = effect x evidence. Effect 1 to 5 is the expected change in Sah's cost per trial start, or the team hours it saves, in the first 4 weeks. Evidence is 3 for a grade A source, 2 for B, 1 for C or D. Evidence grades the source that shows the AI part helps: for a use that saves hours, the page that says what the AI does. A vendor's figure for its own results counts as C. A page or file that shows only the need counts as 1. On a tie, the lower cost ranks first, and a cost not read ranks after every priced row. Then the use Sah can reach sooner, then the group order a, b, c, d, then the name.
| Rank | Score | Use | Group | What it does | Primary source (URL, grade, read date) | Cost (USD, source) | Sah this week |
|---|---|---|---|---|---|---|---|
| 1 | 10 (5x2) | Advantage+ delivery (audience, placements, Andromeda retrieval) | a | Finds the buyers for 3 broad ads inside 4 hard controls. | facebook.com/business/help/938372127764391, A, 2026-10-01 (); effect: Jon Loomer and Ben Heath own data, B | included in Ads Manager | yes: the S-001 plan Decision 2. It learns from Purchase under Core Setup, because Religion limits the data |
| 2 | 9 (3x3) | Meta Ads CLI reads | b | Reads spend and results of the ad account from a script. | developers.facebook.com/documentation/ads-commerce/ads-ai-connectors/ads-cli/setup/get-started, A, 2026-10-06; one source: Meta's own setup page | $0: Meta names no price (same page) | only after an app step: a Meta app, the system user added to it with a role, APP_ID and APP_SECRET in .env. Meta's scope list has ads_management and no ads_read, so the token can write, and the Meta ads guide step 2 starts on reads only. Domen decides the scope |
| 3 | 9 (3x3) | Meta ads MCP reads | b | Gives an agent reports, signal health, help search and the activity log in chat. | developers.facebook.com/documentation/ads-commerce/ads-ai-connectors/ads-mcp-server/ads-mcp-server-overview, A, 2026-10-06 | $0: Meta names no fee (get-started page) | only after Meta's rollout reaches LLC 2 |
| 4 | 6 (2x3) | Meta AI business assistant | a | Gives insights, charts, benchmarks and one-click recommendations in Ads Manager. | facebook.com/business/ai/business-assistant, A, 2026-10-06; one source: Meta's product page, beta | included in Ads Manager | only after it shows in Domen's Ads Manager for LLC 2 |
| 5 | 6 (2x3) | Agent writes through the MCP: paused ads, A/B tests, lift studies | b | Creates and edits campaigns from chat. Ads start paused. | facebook.com/business/help/1456422242197840, A, 2026-10-06 | $0: Meta names no fee | no: writes open later, in stages, on Domen's word (the Meta ads guide step 2). The write block sits on Sah D'Simone LLC (632771467484498), and the LLC 2 guard is unconfirmed |
| 6 | 4 (4x1) | Kajabi join: a script Claude writes once, spend against Kajabi trials by Pacific day | d | Gives the daily cost per trial for the day-8 review and the raise rule. The daily run uses no model. | the S-001 plan, Measurement, A internal for the need, 2026-10-01; no source shows that AI helps | $0 new: Domen's Claude plan, Pro "€18 if billed monthly" (claude.com/pricing, A, 2026-10-06) | yes: the S-001 plan Measurement; Kajabi data in Sah's growth dashboard |
| 7 | 4 (2x2) | Text generation, also text inside images | a | Writes up to 5 primary text and headline variants. | facebook.com/business/help/180641596861873, A, 2026-10-06; effect: Jon Loomer test, B | included in Ads Manager | only after Sah approves each variant and it passes rank 11 (the Meta ads guide step 11; the two constraint files) |
| 8 | 3 (1x3) | pLTV value optimization | a | Bids on a predicted lifetime value. | developers.facebook.com/documentation/ads-commerce/conversions-api/guides/predicted-lifetime-value, A, 2026-10-06 | included in Ads Manager | no: needs "a minimum of 100 conversion events per week" for 4 weeks, and "Datasets in Core Setup cannot use this product"; Religion puts Sah in Core Setup |
| 9 | 3 (1x3) | AI comment moderation and replies (Madgicx AI Marketer) | d | Sorts ad comments, hides negative ones and flags questions. | madgicx.com/blog/can-you-turn-off-comments-on-facebook-ads, A for the feature, 2026-10-06; one source: the vendor's blog | in the Madgicx plan, $45 a month as rendered (rank 21) | no: ManyChat already answers "Comment SUNDAY" for the agency, and a third party would read Sah's account |
| 10 | 3 (1x3) | AI video editing (CapCut: auto captions, AI cuts) | c | Cuts the hook variants and captions from one master. | capcut.com, A for the features, 2026-10-06; one source: the vendor's homepage | not read: capcut.com/pricing answers 404, and the homepage shows no price | only after Sah's team sends the original files |
| 11 | 2 (2x1) | Claude checks every ad text for health and evidence claims | c | Flags claim words before Meta's review or the dataset category sees them. | Need: facebook.com/business/help/2489235377779939 (the Meta ads guide step 10, wording probable, page not rendered); no source shows that a model flags claim words well | $0 new: Domen's Claude plan | yes: Sah's two rules on health and evidence claims |
| 12 | 2 (2x1) | Claude drafts copy and hook variants inside the rules | c | Writes primary text and headlines from Sah's own captions for Sah to approve. | the three-ad brief, A internal for the need, 2026-09-28; no source shows the effect | $0 new: Domen's Claude plan | yes, with Sah's approval and no health claim (the two constraint files) |
| 13 | 2 (2x1) | Claude reads the public Ad Library | d | Collects the angles of running wellness and dance ads before the next concepts. | Need: facebook.com/ads/library/, A for the library, 2026-10-06; no source shows that a model reads it well | $0: public page | yes: no Sah account touch |
| 14 | 2 (2x1) | ManyChat AI steps in the SUNDAY DM flow | d | Lets the DM bot answer free text and send the trial link. | manychat.com/pricing, A, 2026-10-06; one source: the vendor's page | $0 new: Sah is on Pro; list $29 a month (manychat.com/pricing) | only after the agency hands the flows over and the DM link moves to the trial checkout |
| 15 | 2 (1x2) | Generated images and video (image, background, animation, video generation beta) | a | Makes new media from Sah's images. | facebook.com/business/help/1684513971952814 and 1878008236431256, A, 2026-10-06; effect: Jon Loomer, B | included in Ads Manager | no: image generation is not on Advantage+ sales campaigns; every enhancement stays off until Sah approves (the Meta ads guide step 11) |
| 16 | 2 (2x1) | Meta Business Agent on Instagram DMs | d | Answers, qualifies and sells in DMs. | facebook.com/business/news/conversations-2026-introducing-meta-business-agent, A feature, C figures, 2026-10-06 | $0 to start: "Getting started with the Business Agent is free." Then: "In the coming months, businesses will access the agent through our paid subscription offerings" | no: the agency runs Sah's DMs in ManyChat |
| 17 | 2 (1x2) | AI avatar video (Creatify, HeyGen, Arcads) | c | Makes UGC-style ads with synthetic presenters. | creatify.ai/pricing, A, 2026-10-06; arcads.ai, A, 2026-10-06; effect: field study of 384 Facebook ads, B | $39 a month (creatify.ai/pricing); $29 a month (heygen.com/pricing); Arcads not read: /pricing answers 404, the homepage says "Free trial available" | no: the ads are Sah's own reels. The avatar got 38 percent fewer clicks and 13 percent less reach, 25 percent more video plays, and no significant difference in messages (note, Pass 4) |
| 18 | 1 (1x1) | Opportunity score recommendations | a | Scores set-up 0 to 100 and offers fixes. | Meta news post, Meta's page, A feature, C figure, 2026-10-06 | included in Ads Manager | yes, read only: apply none that breaks the S-001 plan Decisions 1 to 4 |
| 19 | 1 (1x1) | Enhancements that edit the video (text improvements, music, subtitles, touch-ups, translations) | a | Changes Sah's media and text per viewer. | facebook.com/business/help/297506218282224, A, 2026-10-06 | included in Ads Manager | no: off until Sah approves (the Meta ads guide step 11); the run is the United States and Canada |
| 20 | 1 (1x1) | Windsor.ai data connector with an MCP | b | Feeds Meta data to Claude. All plans include the Windsor MCP. | windsor.ai/pricing, A, 2026-10-06; one source: the vendor's page | $0: Forever Free plan after a 30-day trial, 1 data source and 1 account; Basic $19 a month (same page) | no: a third party would read Sah's account data, and the CLI reads it with no third party (the Meta ads guide step 2) |
| 21 | 1 (1x1) | AI media buyer (Madgicx, AdAmigo) | b | Builds, tunes and pauses ads by itself. | madgicx.com/pricing and adamigo.ai/pricing, A, 2026-10-06; one source each: the vendor's page | Madgicx $45 a month as rendered: the price depends on a monthly spend tier (Less than $1K to $30K+) and a Monthly, Quarterly or Annual toggle, and the render does not show which is set. AdAmigo $99 a month per ad account under $2,000 a month spend | no: agent writes stay blocked (the Meta ads guide step 2) |
| 22 | 1 (1x1) | AI swipe library (Foreplay) | c | Saves and tags competitor ads. | foreplay.co/pricing, A, 2026-10-06; one source: the vendor's page | $49 a month (same page) | no: the free Ad Library covers 3 concepts (rank 13) |
| 23 | 1 (1x1) | AI traffic routing on landing pages (Unbounce) | d | Sends each visitor to the page variant it predicts converts. | unbounce.com/pricing, A, 2026-10-06; one source: the vendor's page | $249 beside "Billed annually" on the Optimize plan (same page) | no: S-001 starts on one page that copies the home page, and page tests wait until the traffic can read them |
| 24 | 1 (1x1) | AI creative analytics (Motion) | c | Tags creatives and ranks them with an AI analyst. | motionapp.com/pricing, A, 2026-10-06; one source: the vendor's page | $750 a month (same page) | no: sized for up to $50k a month; Sah's week 1 is $252 |
| 25 | 1 (1x1) | AI ad creative generation and scoring (AdCreative.ai) | c | Generates ad images and gives each a predicted performance score. | adcreative.ai, A for the features, 2026-10-06; one source: the vendor's homepage | not read: adcreative.ai/pricing answers 404, and the homepage shows no price | no: the ads are Sah's own reels, and its score is the vendor's own model |
The three uses I would turn on for Thu Oct 8
- Rank 1, Meta's delivery AI. S-001 runs it inside Sah's own engagers: Instagram and Facebook engagers of 365 days, United States and Canada, English, 18 and up, Advantage+ audience off, members and active trials excluded. The AI picks who among them sees each reel. Who: Domen, at the build.
- Rank 11, the claim check. It ranks low on evidence, but it guards two recorded Sah decisions. First step: before the Wed Oct 7 team call, a worker checks every text of the 3 ads in the three-ad brief against the two Sah constraint files and lists each hit. Who: a worker pane, then Sah approves the final text.
- Rank 6, the Kajabi join. It ranks low on evidence, but the $100 funnel check of week 1 and the day-8 review need it from day 1. First step: on Oct 8 a worker pane writes the script (coded trials and trials by Pacific day, team rows out, spend from the ad account read) and runs it daily. Who: a worker pane.
Rank 2, the CLI, needs a token with write scopes. The S-001 reads run on a read-only token instead.
Leave out for now, with the reason
- Ranks 3 and 5, the MCP: Meta's rollout has not reached LLC 2, and writes wait for Domen's stages.
- Ranks 7, 15 and 19, Meta's creative AI: every enhancement stays off until Sah approves it, and generated text can add a claim.
- Rank 8, pLTV: the volume gate and Core Setup block it.
- Ranks 9, 14 and 16, AI in comments and DMs: the agency owns the flows until the hand-off.
- Rank 10, CapCut: it helps the hook cuts after the original files arrive. Oct 8 does not need it.
- Rank 17, avatars: the ads are Sah's own reels, and the one field study shows fewer clicks.
- Rank 18, opportunity score: its fixes push Meta products, so read it and apply nothing that breaks the spec.
- Ranks 20 to 25, paid or third-party tools: each puts a third party in the account, is built for a larger account, or makes media Sah does not need.
Gaps
- No source gives a cost per trial change from any AI use for a creator membership. Each effect is my estimate.
- No source shows that a model checks claim words or reads the Ad Library well (ranks 11 and 13).
- No Meta page names a fee for the MCP or the CLI. The $0 cells rest on that absence.
- Brand memory, the creative approval flow and Muse Image are in testing or unconfirmed (note, Gaps).
- Arcads, AdCreative.ai and CapCut: the guessed /pricing URLs answer 404, and the homepages show no price, so their cost is not read. Automated rules are not a row: Meta's page describes a rule engine, not AI.
- This task made no Meta call, so the business assistant beta on LLC 2 is unchecked.
Three weak points
- The effect scores are judgment for a 3-ad, $36-a-day account. A two-cell test on Sah's account would replace them.
- Meta is the source and the vendor for group a. Its figures (64 percent, 5 or 12 percent) are unaudited, and Haus found Advantage+ ahead in 42 percent of 640 tests (C).
- The practitioner pass is Reddit (D), and the third-party rows rest on vendor pages.