Describe On-Screen Text in the AD Track

Generated spots put the offer in type. The VO says "it just works." The super says "30-day refill · $24 · refill.example.com · cancel anytime." A sighted viewer gets both. An audio-description listener gets the slogan. WCAG's definition of audio description lists on-screen text with actions, characters, and scene changes. If the price lives only in type, and AD never reads it, the described version is a different ad. The craft is to decide which type is information, read it without walking on the VO, and notice when the cleaner fix is to stop putting the offer in type at all.

If the soundtrack already said it, do not say it again

Netflix Audio Description Style Guide v2.5, section 2.1 starts from that check: determine whether the information is already in dialogue before adding it to the description. Legal disclaimers are the exception they name in the same section: read them as-is.

Apply that to a spot:

On-screen	Already in the VO?	AD
"30-day refill" while VO says "a month of product in one bottle"	Paraphrased, not the offer terms	Read the super: "On-screen: 30-day refill."
"$24" while VO never names a price	No	Read the price.
URL in the end card, unspoken	No	Read the URL in a speakable form.
"Cancel anytime" in 8-pt under the CTA	No, and it is a term	Read it.
Product name the VO just spoke	Yes	Skip.
Stylized wordmark the VO named	Yes	Skip, unless the logo animation itself is the gag.
"NEW" bug in the corner for a product the VO introduces as new	Yes	Skip.
Full disclaimer the VO never reads	No	Read as-is, not paraphrased.

"On-screen:" (or Netflix's "words appear") marks that you are reading type. Use it when a naked sentence would sound like a second VO. Drop the cue on a silent end card where everything you say is type.

Paraphrase marketing fluff. Do not paraphrase numbers, dates, percentages, URLs, or legal. "$24" is not "about twenty-five." Netflix's legal-disclaimer rule is the right instinct for regulated lines even off a Netflix timeline.

Generated spots make this worse because models invent type. Placing type by coordinates and editing copy inside a finished graphic keep the picture on approved strings. AD then reads the approved super, not take-three glyphs. If analyze_video returns on-screen text, treat it as candidates. Cross it against the VO and the legal deck before anything is spoken.

Prices, URLs, and the collision with VO

Reading type takes time the hero VO is already using.

Prices. Speak them as a person would: "24 dollars," "from 24 dollars a month." Not "dollar sign two four." A strikethrough next to a sale price is two facts: "was 36 dollars, now 24." If only the sale price is the offer, skip the strikethrough unless it is the point of the shot.

URLs. Strip protocol and www. Say "dot." Split so it can be written down: "refill dot example dot com." Do not read a query string; put a short spoken URL in VO or AD and keep tracking on the destination.

Legal. Read as written. If it cannot fit the gap, do not summarize it as "terms apply." Extend the end card, freeze, or move the full text to a page the VO points to. Compressed legal in AD is a different disclosure than the on-screen block.

Collision rules, in order:

1. If the super lands in a gap, read it in the gap. Standard AD.
2. If the super is on screen during VO but repeats the VO, skip it.
3. If the super is on screen during VO and adds information, you have four moves: slide the super into a gap in the picture; hold the end of the shot after the VO stops and read there; freeze for extended AD; or put the line into the hero VO and delete it from AD.
4. Do not duck the hero VO under an AD read of a super. The hero line is the ad. Talking on top of it makes both tracks worse.
5. End cards with music only are the natural home for URL plus legal. Write the AD for the end card first, then see what still needs a mid-spot read.

The last two seconds of a spot are where this usually belongs, which is why end-frame CTAs should be designed as a silent or near-silent card if you intend to describe them. A VO that talks to 0:29.8 and a URL that appears at 0:28 is a collision you designed.

When a textless-plus-VO version is cleaner

Sometimes the right described version is not AD on the type-heavy cut. It is a cut with the supers removed and those words moved into the soundtrack, so every viewer gets one offer in one place.

Use a textless-plus-VO version when most of these are true:

• The supers are the offer (price, term, URL, legal), not decoration.
• The hero VO currently says none of those facts.
• The cut has no honest gap, so standard AD cannot read the supers without a freeze.
• You would have to freeze several times, turning a 15-second buy into a 25-second described file.
• You already need a textless export for international, publisher, or "no burned-in type" specs.

Then the work is:

1. Export (or regenerate) a textless picture: no supers, no URL card, no legal bug. Keep the product, the face, the motion.
2. Rewrite the hero VO so it speaks the offer: price, term, spoken URL, and any legal that has to be in the file rather than on the page.
3. Record that VO. In Versely that is the ordinary voice-over path: write and generate a voiceover, then add it to the video.
4. Skip AD for those facts, because they are no longer visual-only. You may still need a short AD track for visual beats the new VO does not cover (a pour, a before/after, a gesture).
5. Keep the original type-heavy cut for the placements that want it. Do not overwrite it. The textless-plus-VO cut is the accessible and the localizable sibling, not a replacement thumbnail.

Cleaner than extended AD when the type was doing the VO's job. Worse when the type is decorative, language-specific, or legally required as text. Then you still read the type in AD, as-is, and you keep a type-on picture master.

A practical split:

File	Type on picture	Soundtrack	Who it is for
Picture master	Supers, URL, legal	Hero VO, no AD	Default paid placement
AD master	Same picture	Hero VO + AD that reads supers in gaps or on holds	Blind and low-vision viewers
Textless + VO	No supers	Hero VO rewritten to include price, URL, legal	AD-alternative, localization, publisher specs

You do not always need all three. You need the picture master plus whichever sibling carries the offer to someone who cannot see the type.

A read-through before you mix

Eyes closed, picture master playing, AD script in hand (or the new VO, if you chose textless):

• Can you spend the money? If you cannot, a price is still trapped in type.
• Can you find the product? If the URL is only a QR code or a glyph URL, it is not in the soundtrack.
• Did you contradict the VO? If VO says "from 20 dollars" and the super says "$24," AD must not pick a third number. Fix the picture or the VO; do not "fix" it in AD.
• Did you read type the model hallucinated? If the super is not in the legal deck, it is not in the AD script.

Then mix AD only where the script has a gap or a planned hold. Generate the describer with a reporting delivery, not the hero's energy, and keep it on a copy of the file. The original type-heavy cut stays the original.

FAQ

Do I read every lower-third name and title?

Read them when the listener would otherwise not know who is speaking or why they are on screen, and the soundtrack does not say it. Skip them when the VO already introduced "Maya, head of refill ops" and the lower third repeats it. Credits at the end follow the same gap rule: as time allows, without walking on a closing line.

How do I read a QR code?

You do not describe the pixels of the code. You read the destination if it is also in type, or you add a spoken URL to VO or AD. A QR code with no human-readable URL is visual-only by design; that is a picture problem, not an AD phrasing problem.

Can AD paraphrase a long disclaimer to fit a two-second end card?

Not if the disclaimer is the legal. Netflix's rule is read-as-is; treat regulated copy the same way. If it does not fit, lengthen the card, freeze, move the full text to a page the VO points to, or build the textless-plus-VO version where the disclaimer is spoken at a human rate. A paraphrased "terms apply" in AD is a different disclosure than the on-screen block.

Does burning the super into the picture help AD?

No. Burned-in type is still type. AD cannot OCR it at runtime. Someone has to read it into a script. Burn-in vs sidecar matters for captions, not for description. Description is always a spoken track, written against a list of approved strings.