Skip to content

Lesson 7 of 7 · 6 min read

Check before you post

Run a pre-post check on the final file: claims, AI labels, testimonial words, consent, rights, captions and lips. Then judge hooks on platform data.

A moment at work · fictional example

The ad is rendered and the team likes it. A post goes public in seconds and is hard to take back. Jo wants one check that covers the words, the labels, the rights and the craft before anyone posts it.

You will make: Run the pre-post check on your ad, and plan how you will judge it. Follow the fictional example first, then try it with material you are permitted to use.

Check the file you will post

Run the check on the exported file, not on the script or the edit project. The edit adds text the script never had: stickers, captions, an end card. The post caption is part of the ad too.

Write each line of the check with pass or fail, the evidence and the date. The record shows what you checked if anyone asks later.

Claims and testimonial words

Trace every number and every factual claim, spoken or on screen, to a line in the claim ledger. Each line has a primary source checked the same week, and each number on screen shows its source.

Then read every word for testimonial language. The FTC’s rule on consumer reviews and testimonials, 16 CFR Part 465, took effect on October 21, 2024.

The rule bars a business from writing or creating a testimonial that misrepresents that the speaker exists, used the product, or had the stated experience.

An AI creator is a person who does not exist. She can present, demonstrate and explain. No line can say or imply that she uses the product or got a result from it. Watch for first-person words: “my”, “I use”, “saved me”.

For a line you are unsure of, ask a lawyer before you post.

Disclosure labels

Show the line “AI-generated creator” on screen in the first seconds, at a size a viewer can read on a phone. Then turn on each platform’s own AI label before you post.

TikTok requires a label on AI-generated content that contains realistic images, audio or video. Its help page gives the step for a video: before posting, tap More options, then turn on the AI-generated content setting.

YouTube requires creators to disclose realistic altered or synthetic content. In YouTube Studio, use the AI use setting in the Attributes section during upload.

Meta requires a label on Instagram content with photorealistic video or realistic-sounding audio that was digitally generated or altered. Its examples include a reel narrated with a realistic AI-generated voiceover. On a reel, turn on Add AI label.

Instagram’s help page says labeling works differently for ads. If you pay to promote the video, read the ad platform’s own AI rules as well.

Consent and rights

For a real person on camera, keep a signed release for their face and voice. For any edit or clone of a real person’s likeness, get written consent first. Lesson 6 of the Training videos with AI course covers the state laws.

For an AI creator, keep a record that shows she is not based on a real person: the text description, the reference portrait and the date made. TikTok bars AI-generated content that uses an adult private figure’s likeness without permission.

For every voice, music bed and sound, keep the license terms that allow commercial use, saved with the date read. Sounds you synthesized need only a note of how you made them.

Craft checks

Read the captions against the audio, word for word. Watch mouth crops on P, B and M sounds, where the lips must close. Read the frame grid for hands, faces and text. Confirm the Demo label and every source line are readable on a phone.

Judge it afterwards

Do not judge the ad against a benchmark from a blog. Use each platform’s own numbers for your own posts. For Shorts, YouTube reports Stayed to watch: the percentage of times viewers stayed past the first seconds.

Instagram reports average watch time for a reel: watch time divided by initial views. TikTok’s analytics show each video’s average watch time.

To compare hooks, make a second version that changes only the first seconds. Post both under similar conditions, and compare their numbers with each other. Change one thing at a time, or you cannot tell which change made the difference.

Small audiences give noisy numbers. Compare the two versions after the same number of days, and treat a small gap as no result.

Worked example · Fernway

Jo runs the pre-post check

Jo has the exported MP4, the claim ledger, the words file, the frame grid and the records folder. Fernway will post from its own TikTok and Instagram accounts. Jo runs the check with the file playing on a phone.

  1. Jo lists every claim in the render: “30–40%” of the US food supply wasted, a two-minute check, and “free”. Each traces to a ledger line with its source and the date checked. The slam shows its source line, “Source: USDA estimate, 2010 data”.
  2. Jo reads every word on screen. Over the b-roll of a calm Sunday kitchen sits a sticker the edit added: “My Sunday reset”.That answers the question you guessed at: Guess: the script passed its claim check in Lesson 2. Can the finished video still contain a line that reads as Maya’s own experience?
    What would you do next?The script passed its check. Does the sticker pass?Think it through, then show the answer

    No. “My” makes it Maya’s own routine, a testimonial from a creator who does not exist. Jo changes it to “A calm Sunday”, which shows a picture of a life without claiming anyone lived it.

  3. Jo checks the disclosure. The “AI-generated creator” line shows from the first frame for three seconds, above the stat slam. Jo turns on TikTok’s AI-generated content setting and Instagram’s Add AI label before posting.
  4. Jo checks the records folder: Maya’s text description, her reference portrait and the date made, the stock voice’s license terms, and the music tool’s commercial terms with the date read. The sound effects came from ffmpeg.
  5. Jo checks craft on the phone. The captions match the audio. Mouth crops on “plan”, “bin” and “money” show the lips closing. Every hand in the frame grid has five fingers.
  6. The ad posts. Two weeks later, Jo makes version B, which changes only the hook: Maya asks a question before the stat appears. Jo compares the first week of each version on each platform’s own numbers.
    What would you do next?Version B has a higher average watch time on TikTok and a lower one on Instagram. What should Jo conclude?Think it through, then show the answer

    Nothing firm yet. The gaps are small and the audiences are small. Jo leaves both posts as they are, reads the numbers again after the same number of days, and changes nothing else until one hook leads on both.

The check record lists each line with pass or fail, the evidence and the date. One line failed and was fixed before the post. Jo saves the record beside the video files and starts the next ad from the same ledger.

Go deeper: the research and edge cases

Optional. The exercise needs only what is above.

The FTC rules in detail

Part 465 also covers a business that spreads a testimonial it knew or should have known misrepresented these things. The FTC’s announcement named AI-generated fake reviews as an example of reviews by someone who does not exist.

The FTC’s Endorsement Guides, 16 CFR Part 255, were revised in 2023. They define an endorser as a party who could be or appear to be an individual, group or institution.

Platform labels in detail

TikTok’s Integrity and Authenticity guidelines were released on August 25, 2026 and took effect on September 24, 2026. They require labels on AI-generated or significantly edited content that shows realistic-looking scenes or people.

TikTok may apply its own AI-generated label to content that carries C2PA Content Credentials, and a creator cannot remove that label. Its help page says the setting does not affect distribution if the video follows the Community Guidelines.

YouTube’s list of content to disclose includes a realistic scene that did not occur. Inference: a realistic AI creator speaking in a kitchen fits that description. YouTube may also add a label itself, for example to content with C2PA metadata.

Reading the numbers

On August 24, 2026, YouTube began counting a view when a video starts to play, across all formats. Compare versions measured under the same rule. TikTok’s own analytics tutorial advises watching broad trends rather than one metric.

Starter prompts

  • Run my pre-post check

    Here is everything in my final video ad: the spoken transcript, every piece of on-screen text with its time, the post caption, and my claim ledger: [paste them]. The creator is [a real person, or AI-generated]. Check each item and give pass or fail with the evidence. Check that every number and claim traces to a ledger line with a source. Flag any line in which the creator says or implies she used the product or got a result. Check that an AI disclosure appears on screen in the first seconds. Do not decide legal questions: mark them "ask a lawyer".

  • Find testimonial words

    My video features a creator who [is AI-generated, or is a real person who has not used the product]. Here is all of the spoken and on-screen text: [paste it]. List every line that says or implies the creator used the product, owns it, or got a result from it, including first-person words like "my" and "I use". For each, suggest a rewrite that presents or demonstrates instead. Do not add new claims.

  • Compare two hook versions

    I posted two versions of one ad that differ only in the hook. Here are each platform’s numbers for each version, after the same number of days, with the metric names the platform uses: [paste them]. Compare the versions on each metric. Say where a gap is small enough to be noise, and flag where an audience is small. Do not compare my numbers with outside benchmarks, and do not predict views or sales.

When not to use AI

  • An assistant’s pass alone. A person watches the final file on a phone before it posts.
  • Outside benchmarks for retention or watch time. Compare your own versions on the platform’s own numbers.
  • Posting before the AI labels are on, with a plan to add them later.
Recall from memoryWhy does the pre-post check run on the exported file and not on the locked script?Answer it without looking back, then show the answer

The edit adds text the script never had, such as stickers, captions and an end card. Only the exported file shows everything a viewer sees and hears, so the check reads every word and frame of it.

Exercise · your project

Run the pre-post check on your ad, and plan how you will judge it.

  1. Export the final file and play it on a phone.
  2. Trace every claim to your ledger, and read every word for testimonial language.
  3. Confirm the on-screen AI line, and note the label setting for each platform you post to.
  4. Gather your consent, license and rights records in one folder, with dates.
  5. Write down which platform numbers you will compare, and the one element your second version will change.

Your project workspace

Leave with work you can use.

Write here or use an approved tool. Use fictional or permitted information. Your work stays in this open page and disappears on reload. Nothing is submitted to Emeri or an AI service.

This is self-review. Completion is saved on this device and does not award certification.

Sources