Hooks That Convert: AI Prompt Patterns for Social & Landing Pages (2026)

TL;DR. Great hooks are specific, time-bound, and credibility-forward. Treat LLMs as variant engines: feed audience, offer, proof, banned phrases, and examples of hooks you admire. Never ship raw outputs for YMYL claims. Format outputs with our caption formatter, split long threads using the thread splitter, and study our thirty-day content calendar prompts for batch ideation.

Hooks are not “clever opening lines.” They are compressed arguments that earn the next second of attention. On short video, that is literal: if second zero to one feels generic, viewers swipe. In LinkedIn text, the first line folds in the feed preview. On landing pages, the hero headline must map to the ad click that brought someone there. AI accelerates brainstorming, but without constraints it produces symmetrical fluff: “In today’s fast-paced world…”

This article gives prompt skeletons you can adapt, explains the psychology behind each pattern, and shows how to QA outputs so they stay honest and platform-safe. Pair it with our repurpose video into thirty pieces guide if you are building a repurposing pipeline around the same transcripts.

Anatomy of a converting hook

Specificity beats inspiration

Models default to universals. Fight that by injecting numbers, time windows, locations, and constraints. “Three mistakes” beats “some mistakes.” “Seventeen minutes” beats “quickly.”

Curiosity with a promise

Open loops work when the payoff is credible. Tease a mechanism, not a mystery box. “Why your CPM dropped after the holiday spike” beats “You won’t believe what happened to CPM.”

Social proof without fabrication

Tell the model which proofs are allowed: logged metrics, customer quotes on file, awards you actually won. Prohibit invented testimonials. The FTC cares about deceptive endorsements (FTC Endorsement Guides).

Tone guardrails

Provide a banned phrase list (“unlock,” “level up,” “game-changer”) and examples of on-brand lines you have published before. LLMs imitate local style faster than they invent good taste.

Prompt template A: first-line matrix for Reels or Shorts

Use this when you already have a script body and need ten opening options.

System framing: “You are a senior editor for short video. You write first lines that sound spoken, not like blog SEO. Avoid clichés from this list: …”

User prompt blocks: audience persona, platform (Instagram Reels), niche, offer, three facts from the video, three words we cannot say for legal reasons, desired emotion (urgent, calm authority, playful), and length cap (max twelve words).

Output shape: table with columns: Hook, Why it works, Risk note (if any). Discard any row that invents stats.

After you pick a hook, check safe zones for on-screen text with our social safe-area guide.

Prompt template B: LinkedIn authority opener

LinkedIn favors plain credibility over hype. Feed the model a bulleted resume of outcomes, then ask for hooks that lead with the lesson, not the brag.

Example instruction tail: “No emojis. No rhetorical questions unless they frame a genuine tradeoff. First line must stand alone in preview.”

Cross-read how to grow on LinkedIn in 2026 for cadence and comment strategy so hooks align with distribution.

Prompt template C: landing page hero variants

Give the model visitor context (traffic source, device), primary CTA, and objections from sales calls. Ask for five hero headlines plus subheads, each mapped to an objection.

Run readability checks; for long subheads use reading time & excerpt to keep paragraphs from bloating mobile layouts.

Prompt template D: email subject + preview text pairs

Ask for pairs, not isolated subjects. Preview text should not repeat the subject verbatim. Constrain character counts to your ESP’s typical truncation.

Prompt template E: ad hooks under compliance

For health, finance, or regulated categories, add: “Do not claim cures or guaranteed returns. If unsure, output ‘NEEDS REVIEW’ instead of guessing.” Log those rows for legal.

Prompt template F: TikTok-native pattern interrupts

Pattern interrupts include unexpected visuals, contradictions, and role reversals. Ask the model to propose visual beats tied to each hook so editors can storyboard. See best AI tools for TikTok creators for stack context.

QA checklist before publishing

1. Fact check numbers and names.
2. Read aloud for mouth feel.
3. Remove duplicate meaning between hook and on-screen text.
4. Check claims against source docs.
5. Platform policy scan for prohibited content in your niche.
6. Accessibility: contrast for burned-in captions via contrast checker.
7. Hashtag hygiene if posting with tags: hashtag normalizer.

Why “just be more creative” fails

Creativity without constraints produces variance, not quality. Constraints channel search through a smaller space of acceptable phrases. That is why providing three gold-standard hooks you wrote beats providing a twenty-page brand deck with no examples.

Batching workflow for teams

Monday: model generates twenty hooks from a shared prompt library. Tuesday: humans shortlist six. Wednesday: film two variants per shortlisted hook for high-stakes campaigns. Thursday: publish; Friday: retro metrics (watch time, CTR, saves). Store winning prompts in Git or Notion with metadata: audience, offer, season.

Ethics and disclosure

If content is AI-assisted, follow client and platform disclosure norms. Authenticity is not “no AI”; it is no deception. For sponsored posts, disclosures must be clear and proximate (FTC guidance).

Connecting hooks to bio and UTMs

Hooks earn attention; bios and landing pages capture intent. Align CTA language across surfaces. Use the bio character counter and UTM builder so traffic traces to the hook variant.

When to avoid LLMs entirely

Crisis comms, individualized legal/medical advice, and sensitive HR announcements should be human-led. Models can still help with internal brainstorming, not external copy.

Advanced: chain-of-thought without leaking to users

Some teams ask models for private reasoning first, then a public-facing hook. If you use this pattern, strip hidden reasoning before publishing. Never paste chain-of-thought into CMS fields by accident.

Prompt template G: “negative space” hooks

Negative space means saying what something is not before what it is. Example shape: “This isn’t another tips thread. It’s the checklist we used before spending six figures on the wrong agency.” Feed the model two forbidden framings and the true framing you want to defend. This pattern works when audiences are cynical about category noise.

Prompt template H: founder story without humble-brag poison

Ask for scene-setting with sensory detail, then stakes, then lesson. Constrain adjectives; require one verifiable fact per paragraph (date, metric, customer count band). If the model inflates numbers, your instruction should say: “If no exact metric is supplied in context, write ‘undisclosed’ rather than inventing.”

Prompt template I: carousel slide one as hook

Carousels live or die on slide one. Provide the full outline and ask for five first-slide sentences plus suggested on-slide typography (max six words). Then sanity-check legibility with the contrast checker once your designer picks colors.

Prompt template J: paid social under character caps

Give the model exact character budgets for Meta primary text, headline, and description fields. Ask for a table that counts characters explicitly. If the model miscounts, rerun with “verify counts twice.”

Style transfer without plagiarism

Paste links to your own past posts (not other people’s copyrighted text) as style references. Ask the model to imitate structure and rhythm, not wording. For competitor inspiration, describe patterns in your own words instead of dumping their copy wholesale.

Governance for agencies

Maintain a prompt library with owners, version tags, and change logs. When a client churns, archive their banned words and compliance notes so they do not leak into another account. For multi-brand studios, physically separate workspaces in your LLM vendor if available.

Measuring hook performance beyond vanity

Pair platform metrics with site events when hooks push outbound clicks. Clean URLs with the link cleaner when recycling landing pages across campaigns so old UTM parameters do not pollute new reports. For revenue conversations, align hooks with the same vocabulary finance uses in CRM stages; our CRM SMB guide helps keep language consistent from ad to close.

Localization: hooks that survive translation

If you localize, avoid idioms that require footnotes. Ask the model for a literalness score per hook (its self-assessment, not ground truth) and still have native speakers review. Numbers and currencies must match the market; miles vs kilometers jokes do not travel. Keep a glossary of product names that must remain untranslated.

Seasonality and fatigue

Holiday hooks exhaust audiences fast. Maintain a seasonal calendar with “retired phrases” so nobody ships the same Thanksgiving guilt line clients saw last year. Rotate emotional registers: gratitude, humor, blunt utility. Models will repeat seasonal tropes unless you explicitly ban last year’s list.

Micro-editing pass humans should always run

Even strong model outputs benefit from five minutes of human polish: swap weak verbs, remove redundant intensifiers, tighten numbers, and align the hook’s promise with the first payoff moment in the body. If payoff arrives too late, either move content or rewrite the hook so expectations stay honest. Small honesty upgrades compound trust over quarters, not single posts. Keep that standard visible in your creative briefs.

FAQ

How many hook variants should I generate?

Ten to twenty internally, ship two to three publicly for testing.

Which model is best?

The best model is the one your team can prompt consistently with logging. Compare on your own prompts monthly; capabilities shift.

How do I stop robotic tone?

Ban corporate phrases, require contractions or forbid them uniformly, and add “write like a specific magazine.”

Can I paste customer reviews into prompts?

Yes if you have rights and redact PII. Do not fabricate amalgamated quotes.

Do hooks differ by industry?

Yes. B2B often rewards specificity; consumer beauty rewards sensory language. Tune examples accordingly.

Should hashtags appear in the hook sentence?

Usually no; keep hooks human-spoken and put tags below or in first comment depending on strategy.

How long should a Reels hook be?

Short enough to speak in under two seconds unless deliberate slow burn.

What is a simple A/B test?

Same body, different first line; compare average percent viewed.

References

1. OpenAI Documentation — API and prompting basics.
2. Anthropic Claude Docs — long-context prompting patterns.
3. Google AI for Developers — Gemini developer guides.
4. FTC Endorsement Guides — truthful advertising.
5. APA style on inclusive language — thoughtful language defaults.
6. Nielsen Norman Group — readability and web writing research.
7. Meta for Creators — platform-native guidance.
8. LinkedIn Marketing Solutions — B2B creative context.
9. YouTube Creator Academy — retention and packaging.
10. W3C WCAG — accessible contrast and structure.
11. American Marketing Association — ethics and professional standards.
12. EFF on platform speech policies — context for moderation debates.
13. Harvard Business Review — credible framing for leadership hooks.
14. APA PsycNet — primary literature gateway for persuasion research.
15. IEEE publications — technical depth when writing for engineers.