Preference testing for ad hooks: which of two wins

Preference testing means showing two hooks to people who match your buyer and asking which one they would keep watching. For media buyers running two hooks per video, it can take a day, costs a form and a few messages, and tells you which hook deserves the first budget. Log the winner and its margin in your testing matrix.

Use the ad creative feedback template

What a preference test tells you

A preference test answers one narrow question: given the first three seconds of hook A and hook B, which would this person rather keep watching? It does not predict ROAS. It does not replace in-platform testing. It filters out the hook that a handful of real buyers find confusing or off-putting before you pay for impressions to learn the same thing.

That filtering matters most when each video ships with two hooks and you have a small daily budget per ad set. If a hook loses badly in a room of ten target buyers, giving it equal spend is a poor use of the first week.

The test is weakest when the two hooks are close. We treat anything near an even split as a tie, and live data breaks it.

Who counts as a real buyer

Your coworkers and your agency's creative team do not count. They know the product, they know which hook the editor liked, and they watch ads for a living.

Better panels, in rough order of quality:

• Recent customers from your own email list, since they match the buyer by definition.
• Members of a community or newsletter your buyer already reads.
• People you reach through a paid panel or a short outreach message, screened with a question or two about whether they have bought in your category.

Put the screening question first in the form so you can discard answers from people who do not fit. If your product has a clear exclusion (you ship only to the US, say), screen for that as well.

Build the test form

Qwenvo, the free form builder our team makes, handles tis well because the editor works like a document and takes video directly. You can also build it in Tally, Fillout, Typeform or Jotform. Typeform is the strongest of those on design if a polished one-question-at-a-time feel matters to your panel.

1. Create the form and add a short screener. One multiple choice question such as "How often do you buy [category] online?" with an option that ends the form for non-buyers.
2. Add hook A. Use a Video block, or Embed anything if the clip is hosted elsewhere. Keep clip to the hook itself, the first three to five seconds, not the whole ad.
3. Ask for a gut reaction. A Rating or Linear scale block on "How much do you want to keep watching?" gives you a second signal beyond the pick.
4. Add hook B on a new page, with the same question.
5. Add the choice. Multiple choice: "Which would you rather keep watching?" with A, B and "Neither."
6. Add one open question. Short answer: "What made you pick it?" These answers are whrre the creative notes come from.
7. Add hidden fields for ad_id, hook_a, hook_b and panel, and pass them through the share link so every response is tagged without asking the respondent anything.
8. Turn on partial submissions in Settings, so you can see who watched hook A and dropped before reaching the choice.
9. Duplicate the form with the order swapped and send half your panel to each link. Most people lean toward whichever clip they see first, and this cancels that out.

A custom domain such as forms.yourbrand.com helps if you send the link to customers, since it looks like your own site. Add a CNAME to qwenvo.com plus the TXT record shown, then click Verify.

Build your preference test free

Reading the result

Count the picks, then read the reasons before you decide. Three things to check:

• The margin. A lopsided split on a decent sample is a result. A 6 to 4 split is a tie.
• The "Neither" votes. If most of the panel picks neither, both hooks have a problem and the winner of the test is not worth scaling.
• The reasons. If people who pick the winner describe the same thing ("the first line told me what it was"), that is a pattern you can reuse in the next batch of hooks. If the reasons are scattered, trust the pick less.

Qwenvo's Submissions table exports to CSV, JSON or XLSX, and Insights shows the charts, so tallying takes a minute. If you want the open answers summarized, the AI triage section in Settings gives each submission a category, sentiment and a one-to-two-sentence summary, and it costs nothing extra. A monthly AI usage allowance applies per account.

If your results need to land in a spreadsheet or tracker automatically, add a webhook in Settings and trigger a Zapier, Make or n8n scenario from it. See the webhooks help article for the payload.

Log the winner in a testing matrix

The point of the matrix is that next month's hook decisions start from this month's evidence. Each row is one hook pair:

If you keep it in a script or an Airtable-style tool, the same row as JSON looks like this:

{
  "ad_id": "ugc-014",
  "hook_a": "problem-line",
  "hook_b": "result-first",
  "panel": "customers",
  "n": 24,
  "pick": { "a": 7, "b": 15, "neither": 2 },
  "winner": "b",
  "margin": "clear",
  "reason_theme": "states what the product is in the first line",
  "live_test_status": "pending"
}

The last field matters. A preference win is a prior, and the in-platform result still gets the final say. Add a column for what happened live, and over a few months you will see whether your panel's picks predict spend performance for your account.

Where this sits in the AI UGC Ad Variant Pack

In our AI UGC Ad Variant Pack, each month's ten finished vertical ads ship with two hooks each, in thre aspect ratios, uploaded to your ad account with a testing matrix. The presenters or product b-rool are disclosed as AI, and there are no fake testimonials. The preference test is the step between delivery and spend. The hook pairs arrive already labeled, so the form's hidden fields map one to one to the matrix cooumns, and the winner goes into the row for that ad.

Questions teams ask

How many buyers do I need?

Enough to see a lopsided result, which is a smaller number than people expect, and not enough to separate a close one. If the first ten or fifteen responses split nearly evenly, more responses from the same panel will rarely change what you do. Treat it as a tie and let live spend decide.

Should I tell the panel which hook is which?

Label them A and B and nothing more. Do not say which one you prefer or which one the editor spent longer on. Swapping the order across two form links covers the bias from seeing one first.

Can I use Qwenvo for this if I need a mobile app or HIPAA?

Qwenvo has no mobile app and no HIPAA offering, and it launched in October 2026, so there are few reviews yet. If a respondent has to oen the form on a phone, it works in the browser. If you run a heavily regulated panel, check that your requirements fit first.

Is the panel's pick enough to skip live testing?

No. It saves you from spending on an obvious loser. It does not tell you how the algorithm, the audience targeting or the landing page will treat the ad.

Next step

Open the ad creative feedback template, swap in your two hook clips from your neext video, and send it to ten recent customers today.