Write the 16-Character Headline First, Then Earn the Other 34
By Keshav Parsai 8 min read
In brief
Published ChatGPT Ads headline limits conflict. Use a shortest-first writing method that remains clear under every reported limit.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 October 2026
Five published sources give ChatGPT Ads headline limits. They do not agree. The range runs from 16 characters at the low end to 50 at the high end, with 32 sitting in the middle, and no source cites an OpenAI specification page that settles it.
Our specs reference tracks that conflict and keeps it current. This article assumes you have read it and answers the next question: how do you write a headline when you do not know how much of it will survive?
The rule the conflict forces
Write the headline that works at 16 characters. Then extend it, and treat every character past 16 as something you would be content to lose.
This is not a compromise position. It is the only position that is correct under every published limit at once. A headline built at 50 and truncated to 16 becomes a fragment. A headline built at 16 and extended to 50 degrades into the thing it started as, which was already a complete claim.
Truncation is documented on this format. Length limits are contested. When one of those is certain and the other is not, design against the certain one.
One offer, three lengths
The offer, held constant: UK payroll software for small employers, files RTI submissions automatically, from 29 pounds a month.
At 16 characters or fewer
| Headline | Chars | What it does |
|---|---|---|
| RTI filed free | 14 | Names the task and the hook. Weakest on credibility |
| Never miss RTI | 14 | Names the fear. Strongest for a reader who has missed one |
| Payroll in 1hr | 14 | Names the time saving. Vague about which hour |
| Payroll, sorted | 15 | Register only. Says nothing checkable |
| £29 payroll | 11 | Price qualifier. Filters hardest, converts narrowest |
At this length you get one idea and one only. Notice that "Payroll, sorted" is the weakest of the five despite reading most naturally, because it is the only one that carries no fact. When space is this short, tone is the first thing to cut, not the last.
At 24 characters or fewer
| Headline | Chars | What 24 bought over 16 |
|---|---|---|
| RTI filed automatically | 23 | The mechanism. "Automatically" is the whole differentiator |
| Payroll from £29 a month | 24 | A qualified price rather than a bare number |
| Never miss RTI deadlines | 24 | Plural, which implies ongoing rather than one rescue |
| Payroll done in 20 mins | 23 | A specific number instead of a round hour |
Twenty-four characters buys exactly one modifier. Spend it on the word that makes the claim checkable: "automatically", "from", "20". Do not spend it on an adjective.
At 50 characters or fewer
| Headline | Chars | What 50 bought over 24 |
|---|---|---|
| UK payroll software that files RTI automatically | 48 | Geography and category, so the wrong reader self-selects out |
| Payroll for teams of 1 to 20, from £29 a month | 46 | Size qualifier plus price, two disqualifiers in one line |
| Missed an RTI deadline? File the next one on time | 49 | A trigger event and a resolution, which is the strongest form |
Fifty characters is enough for a trigger and a consequence, which is the only structure in this set that describes a situation rather than a product. That is also the version that loses most under truncation, which is the trade.
Front-loading, and where the cut lands
Assume the cut. Then ask what the first 16 characters of each 50-character headline say on their own.
- "UK payroll software that files RTI automatically" cuts to UK payroll softw, which is a category with no claim.
- "Payroll for teams of 1 to 20, from £29 a month" cuts to Payroll for team, which is a category with no claim.
- "Missed an RTI deadline? File the next one on time" cuts to Missed an RTI de, which is a broken question.
All three fail, and they fail in the same way: the load-bearing word sits past the cut. Rewrite so it does not.
- "RTI filed automatically for UK small employers" cuts to RTI filed automa, which still reads as the claim.
- "£29 a month payroll, RTI filed for you" cuts to £29 a month payr, which still reads as an offer with a price.
The discipline is one line long: the first 16 characters must be a sentence you would be willing to run alone. Everything after is qualification.
Counting characters properly
Three counting mistakes cost real characters.
The pound sign, the euro sign and curly quotes are single characters in most counters and two or three bytes in some. If a limit is enforced in bytes rather than characters, "£29" costs you an extra character you did not budget for. Nobody has published which ChatGPT Ads uses. Assume the pessimistic version.
Spaces count. Obvious, and still the most common reason a headline pasted from a document comes back rejected at the edge.
Trailing spaces and non-breaking spaces survive copy-paste. Paste into a plain counter, not into a word processor.
The reckoning: this may be over-engineering
If the real limit is 50 and stays 50, everything above is a constraint you imposed on yourself for nothing, and your headlines are shorter and blunter than the format allows. That is a genuine cost. Short headlines carry fewer disqualifiers, and on a channel with no exclusion layer, disqualifiers in the creative are one of the few filters available.
The defence is asymmetry rather than certainty. Writing short when long was allowed costs you some qualification. Writing long when short was enforced costs you a headline that reads as a fragment next to an answer the user trusts, in a placement where looking broken is worse than looking plain.
We cannot size either cost. No advertiser has published truncation rates, and OpenAI does not report whether an impression was served truncated.
What we cannot tell you
- The actual headline character limit. Five published sources conflict, from 16/32 to 50/100, and OpenAI has not published a specification page that resolves it.
- Whether the limit is counted in characters or bytes. Undocumented, which matters for currency symbols and accented characters.
- Whether limits differ by surface. The format renders on iOS app, Android app and desktop web, and nobody has published per-surface truncation behaviour.
- How often your ads were truncated. Not reported at any level in Ads Manager.
- Whether truncated headlines underperform. No published test, and InPromptAds has no first-party data.
Quick answers
What is the ChatGPT Ads headline character limit? Contested. Published sources give figures from 16 to 50 characters, and OpenAI has not published a specification that settles it. Treat the conflict itself as the working fact.
How long should I write my headline? Write a complete claim in 16 characters, then extend to the longest limit you believe applies. The first 16 characters must stand alone as a sentence you would run by itself.
What should I cut first when shortening a headline? Tone and adjectives. Keep the one word that makes the claim checkable: a number, a price, a mechanism, a deadline.
Does truncation happen in ChatGPT Ads? Yes, truncation on the ad format is documented. What is not documented is where the cut lands for a given surface or how often it happens to your ads.
Should I include a price in a short headline? Only if filtering matters more than volume. A price is the strongest disqualifier available in 11 characters, and it will lower click-through by design.
Do currency symbols cost extra characters? Possibly. If the limit is enforced in bytes, the pound and euro signs cost more than one. It is undocumented, so budget for the worse case.
Sources
| Claim | Source | Tier |
|---|---|---|
| Ad creative is brand name, logo, headline, description, image and destination URL | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Headline character limits published between 16/32 and 50/100 across five sources | Lapis, LocalAds, Barefoot Performance, IOquery, AdSpyder ChatGPT ads specs write-ups, 2026 | Reported, conflicting |
| Truncation occurs on the ad format | IOquery, ChatGPT Ads chat card format and truncation limits, 2026 | Reported |
| No truncation reporting exists in Ads Manager | OpenAI Ads Manager documentation, by absence | Absent |
| Write at 16 first, front-load the claim, treat extension as expendable | InPromptAds reasoning from the documented truncation plus contested limits | Inference, ours |
Related reading
- ChatGPT Ads Specs
- ChatGPT Ad Format
- Writing ChatGPT Ads That Read as Useful Rather Than Promotional
- The 256px Square: What Actually Reads at ChatGPT Ad Image Size
- ChatGPT Ads Checklist
Changelog
12 September 2026, v1.0. First publication. Establishes the write-at-16-first discipline and the front-loading rule for surviving documented truncation under contested limits.
Field kit
Tools
Site
OpenAI (primary)
Keshav studies how AI systems retrieve, verify, and cite brand information. At InPromptAds, he leads source research and turns platform documentation into practical guidance for advertisers.
