What is HOOK from Nagent?
HOOK is Nagent's AI creative strategy lead for paid social, and it is in Beta. It turns a brief into ad angles, hooks and scripts, checks every claim in them against the proof your brand has on file, and packs the result for a person to produce and upload. It also reads how your ads performed, calls winners and flags fatigue. It never launches, pauses or changes the budget of a live ad. HOOK sits on Nagent's Marketing team and reports to MIRA.
What does HOOK actually do?
HOOK is not an ad generator and not a media buyer. It is the strategist in the middle: it decides what to say and why, makes sure you can prove it, and learns from what ran. The full role is on the HOOK page. It runs one loop:
Brand pack first. Voice, buyer profile, the claims the brand may make with their proof, banned language, the offer and the channel packs. Nothing is made until the pack is complete.
Research. It builds a swipe library of competitor and category ads worth studying, with the reason each was kept, and tears video ads down into their beats.
Angles. It turns a pattern it found into a testable angle for your brand, with the claim the angle rests on written out.
Hooks and scripts. It writes hook sets per concept and channel pack. A person pins the sets they like and hands them on for full scripts and shot lists.
The gate. Every claim and line of copy is checked against the brand pack. Anything the pack does not support is held.
Ship. Statics go to RUPA, Nagent's image studio, and video briefs go to Alpha, its video studio. Approved work is packed into one handover a producer can upload by hand.
Learn. It reads ad-level results, ranks each ad against your own account's ads, diagnoses losers, writes a refresh plan and keeps a playbook of what won.
The channel packs today are Meta cold (9:16), Meta retargeting (1:1), TikTok Spark and Google responsive search ads. Each carries fixed format limits, so hooks and copy are written at the length they will run.
What are HOOK's eight sub-agents?
HOOK leads eight seats, all in Beta, each with one job:
SCOUT (Ad Scout): finds competitor and category ads worth studying and says why each earned a place in the swipe library.
MINE (Pattern Miner): works out what makes an ad work and states it as a mechanism, not a screenshot.
ANGLE (Angle Builder): turns a mined pattern into a testable angle, writes the brief and gives every bet a kill condition.
SCRIPT (Script Writer): writes hooks, scripts and shot lists a producer can shoot without a briefing call, and checks every claim a script cites.
GATE (Claim Gate): checks claims and copy against the brand pack and holds what the pack does not support. It is the only seat that stops work rather than producing it.
PACK (Campaign Packer): assembles one handover carrying the angle, the script and the constraints together.
PULSE (Performance Reader): reads ad performance and separates a weak angle from a weak execution of a good one.
VAULT (Prior Keeper): keeps what each campaign taught, so the next brief starts from it.
What does HOOK need from you before it makes anything?
A complete brand pack. It has four steps: voice and buyer profile, claims and proof, banned language and offer, and channel packs. The pack counts as complete when it has a voice, a buyer, at least one claim and at least one channel pack.
The claims step is the one to take seriously. Each claim you allow in an ad is saved with its proof. GATE treats any number, percentage, price or guarantee word in a line ("guaranteed", "risk free", "clinically", "100%" and similar) as a claim until a live claim or your offer backs it. That mirrors what advertising regulators already expect. The UK advertising code, for example, says marketers "must hold documentary evidence" for objective claims before an ad goes out (CAP Code, rule 3.7).
Voice and banned words are read live from the workspace's brand settings, so an edit there reaches every new brief, hook and script.
How does HOOK decide an ad won or is fatiguing?
This is where HOOK is more specific than most tools, because the rules are fixed in code rather than left to a model's judgement:
Winners are called against your own ads. Each ad is ranked against this account's own ads in its channel pack over a 90-day window. An ad needs at least 1,000 impressions to be ranked. A video ad is a winner when its hook rate and hold rate both sit at or above the 75th percentile; a static needs its click-through rate there.
Small accounts get a hunch, not a verdict. With fewer than 30 ranked ads in the population, HOOK shows the value but makes no win or lose call.
Fatigue needs two weeks of delivery. An ad is flagged as fatiguing only after 14 delivery days, when its last seven days' hook rate (video) or click-through rate (static) is 25% or more below its first seven days, with at least 1,000 impressions in each span.
Frequency is a refresh reason. A seven-day frequency of 3.0 or more is listed as a reason to refresh.
Kill conditions are set before launch. Each bet in a brief carries a structured kill condition: a metric (hook rate, hold rate, click-through rate or cost per acquisition) below a percentile after a set spend or number of days. PULSE checks it on every read and says when one is met.
With a Meta account connected, reads and diagnoses run nightly and the refresh plan and playbook weekly. Without one, you upload a .csv or .xlsx export from Meta Ads Manager, Google Ads, TikTok Ads Manager or another platform, and HOOK reads it when you save the file or press Read now.
What can HOOK not do?
Plenty, by design. Today:
It makes read calls only on an ad account. It cannot launch, pause, or change a budget or a bid, and none of its sub-agents can. When PULSE says a kill condition is met, a person pauses the ad in Ads Manager.
Meta is read only. HOOK cannot place Meta campaigns; you download the package and upload it yourself.
Google packs go through NIA. HOOK hands a gated Google responsive search ad pack to NIA, Nagent's performance marketing lead, who builds the campaign paused and asks for her own approval first. Google Ads results are not read back into HOOK yet.
TikTok packs cannot be placed, and no TikTok video is gated yet.
A gated video cannot go into a pack yet. Packs hold statics, carousels and Google responsive search ad sets.
Chat is lighter than the full workflow. Talking to HOOK in chat uses a general prompt, not its full workflow; the structured work runs from HOOK's own pages.
What waits for a person's approval?
Drafting, checking and reading run on their own, because they cost little and change nothing outside your workspace. Paid research waits: an ad library search and a teardown both wait for approval, and so does sending a brief to RUPA, because it spends RUPA's budget. Overriding a GATE block always needs a second person: the person who asked for the override cannot approve it.
A workspace admin sets how much HOOK and each seat may do alone, and every change lands in the audit log. Across Nagent, agents arrive at L2 Assist on a five-level ladder and are promoted one step at a time by a person; the autonomy and governance docs explain the ladder.
How is HOOK different from AI ad generators and creative analytics tools?
Most tools in this space do one side of the loop well. We make Nagent, so here is a fair read of the five we get compared with most, from their own sites, checked 24 October 2026:
AdCreative.ai: generates ad creatives and copy, with Creative Insights that it says monitors connected ad accounts "to detect signs of ad fatigue", and a Compliance Checker for brand, platform and policy rules. Pricing on its site starts at $39 a month for Starter (20 credits, one brand) and $249 a month for Professional, billed monthly. Best for teams that need finished designs fast.
Pencil: an AI platform that orchestrates several image, video and text models to generate, edit, test and launch ads, with brand safety guardrails and governance controls. Its own summary lists Core at $14 a month, Growth at $55 a month and Pro by quote. Best for enterprise teams producing high volumes across markets.
Omneky: generates image, video and avatar ads from a brand kit read off your website, launches them on Meta, Google, TikTok and other channels, and scores creatives in Insights. Plans run from Lite at $29 a month to Pro at $249 a month, with Enterprise by quote. Best for brands that want generation and launch in one place.
Motion: creative analytics for paid social, priced by monthly ad spend; Starter is $999 a month for up to $50k of monthly spend. Best for teams that already produce plenty and need to read results well.
Foreplay: an ad library, swipe file, competitor tracking (Spyder), creative analytics (Lens) and a brief and script builder (Briefs). Basic is $59 a month, Workflow $175. Best for strategists building research and briefs.
HOOK's difference is the claim gate and the rules. AdCreative.ai checks brand, platform and policy rules, and Pencil says it enforces brand guidelines and legal compliance; neither page we read describes checking each script line against a stored list of your own claims and their proof. HOOK, for its part, does not generate finished visuals on its own; it hands statics to RUPA and video to Alpha. If you need volume of finished assets this week, a generator fits better. If your problem is that nobody can say why an ad won or whether a line is provable, HOOK is built for that.
Who is HOOK for?
The paid social manager and the creative strategist, not instead of them. HOOK hands the team the angle, hooks, scripts and constraints in one pack, and people decide what runs. It suits:
brands in regulated or claim-sensitive categories, such as health, finance or skincare, where one unprovable line costs more than a weak ad;
teams whose creative fatigues faster than they can brief, who need a steady supply of angles tied to evidence;
teams that run ads on Meta but keep their results in exports, since HOOK can learn from uploads alone.
It is a poor fit if you want a tool that launches and optimises ads on its own. That is not HOOK's job, and on Meta it is not something Nagent does without a person either.
How much does HOOK cost?
HOOK works on the Marketing team, so it comes with the plan that includes that team; the pricing page shows which. All agent work in Nagent is metered in credits, visible in the workspace. For scale, a campaign build is about 40 credits and a rendered video about 700. HOOK shows its own spend on its Today page, and its "cost per winner" counts HOOK's own spend only, never your ad spend. If a workspace passes its daily limit, HOOK stops all work until the limit resets.
Frequently Asked Questions
What is an AI creative strategist for paid social?
It is an agent that decides what an ad should say and why, then learns from results, rather than only producing assets. Our longer explainer on what an AI creative strategist does covers the role in full. HOOK is Nagent's version, in Beta.
Does HOOK launch or pause ads on its own?
No. HOOK makes read calls only on an ad account. A person uploads, pauses and spends in the ad platform. For Google, PACK hands a pack to NIA, who builds it paused and waits for approval.
Which ad platforms can HOOK learn from?
A connected Meta account, read nightly, or a .csv or .xlsx export from any ad platform, including Google Ads and TikTok Ads Manager. The uploaded file itself is not stored.
How is HOOK different from AdCreative.ai?
AdCreative.ai generates finished creatives and scores them. HOOK writes the strategy, checks claims against proof and reads results, then hands production to RUPA or Alpha. We compare them in detail in HOOK vs AdCreative.ai.
Does HOOK replace our creative team?
No. A person pins the hooks before SCRIPT writes full scripts, a person approves paid research, and a person ships. HOOK removes the research, checking and reporting work around the team.
HOOK is in Beta on the Marketing team; see it on the HOOK page, or book a demo and we'll walk you through a brand pack.