One Deck, Three Audiences: Rework, Do Not Rebuild

To adapt a presentation for different audiences with AI, keep one source of content and produce cuts from it, rather than generating three decks and maintaining three decks. What changes between the board, the delivery team and a wider internal audience is the order, the depth and the framing. What must not change is the numbers, the commitments and the ask. A generation tool is good at the first list and dangerous on the second, because it will happily rephrase a figure or soften a commitment while it is adjusting tone. So the working rule is simple: let the tool rework the shape, and move every number through a single source you control.

What changes between audiences, and what must not

The advice you will find across this topic is broadly right and worth stating plainly. Decision-makers want the outcome, the cost, the risk and the decision they are being asked for, with detail available but not in the way. The team that has to execute wants sequencing, dependencies, who does what and what is unresolved. A wider internal audience wants context and what it means for them, and needs less prior knowledge assumed.

So four things legitimately change between cuts:

• Order. What is on slide one differs by audience, and it is usually the biggest single difference.
• Depth. The same point gets one line in one cut and a slide in another.
• Framing. Outcome-first for a decision audience, process-first for a delivery audience.
• Vocabulary. Internal shorthand that is efficient with your team is noise to anyone else.

And four things must be identical in every cut:

• The numbers. Same figures, same basis, same period.
• The ask. If you are requesting a decision, it is the same decision in every version.
• The commitments. Dates and deliverables do not get more optimistic for the audience who wants to hear that.
• The risks. Reordering a risk is fine. Dropping it from the version that goes to the people who could act on it is not.

That second list is the whole reason this article exists, because a rewriting pass does not know which is which. A tool asked to make a deck "more executive" will shorten, and shortening is exactly the operation that removes a qualifier from a number or drops the caveat attached to a date.

The master and cuts model

Build one master, then derive.

The master is the complete version: every slide, every number, every caveat, in a logical order. It is not the deck you present to anyone. It is the source. Then each cut is a selection and a reordering of the master, with a new opening slide written for that audience.

Three practical properties make this work rather than becoming three decks by another name.

Slides move, they do not get retyped. A cut is made by hiding, deleting or reordering master slides in a copy, not by rewriting the same content again. Retyping is where versions diverge.

Detail goes to an appendix rather than being deleted. The decision audience gets a short main sequence and the full detail behind it, which means one file can serve a live presentation and the follow-up question at the same time.

One cut is nominated as the source of truth. When a number changes, it changes in the master first and propagates. Nothing gets corrected only in the version that happened to be open.

Where the numbers live

The most expensive failure in multi-audience decks is not tone. It is the same metric appearing with two different values in two versions, discovered by the one person who has seen both.

Keep the figures in a single place outside the slides, most often the spreadsheet or report they came from, and pull them into every cut from there rather than typing them per version. If your charts are generated from that data, regenerate them from the same source for every cut instead of copying an image between files, and check them, because charts built from spreadsheet data have specific and repeatable ways of going wrong that survive a copy and paste unchanged.

The same logic applies upstream. If the content originally came from a report or a memo, that document is the natural master, and generating each cut by turning that document into slides with different instructions about depth keeps a single written source behind all three decks. This is more robust than deriving deck from deck from deck.

What a tool can safely do here

Editing features in this category are documented to operate on existing content, which is what a cut needs. Gamma's help center describes an agent that can make edits across all slides at once for consistent updates as well as editing individual slides, including summarization, readability improvements and custom instruction-based editing, and notes that after the agent makes edits you can refine them manually (Gamma Help Center, checked 2026-09-07). Plus AI documents rewriting existing slide text to condense, expand or change tone, and reformatting an existing slide into a different shape (Plus AI, checked 2026-09-07).

Those are the right operations for producing a cut: condense this section, retone this deck for a non-technical audience, summarize these four slides into one. They are also the operations most likely to quietly remove a caveat, so each one is followed by a check rather than trusted.

Two things to keep out of the tool's hands entirely. First, the numbers: never ask for a version where figures are "simplified" or "rounded for the executive audience", because a rounded number in one cut and an exact one in another is a discrepancy waiting to be found. Second, the opening slide of each cut: it carries the claim that version rests on, and it is worth the five minutes it takes to write yourself.

The three cuts in practice

The decision cut. Opens with the recommendation or the request, then the two or three numbers that support it, then risk, then what you need and by when. Everything else is appendix. Length is short enough to reach the ask early. The single most common defect is burying the ask on the last slide.

The delivery cut. Opens with what is changing and what it means for the work. Carries sequencing, dependencies, owners and open questions. The numbers appear where they affect a decision someone in the room has to make. Detail that a decision audience would find excessive is exactly what this audience came for.

The wider cut. Opens with context, because you cannot assume prior knowledge. Uses fewer internal abbreviations, more explanation, and states what it means for the reader specifically. This is usually the version that gets read alone rather than presented, which means the slides have to stand without a speaker, and the speaker notes matter less than the slide text.

For the cuts you do present live, each version needs its own notes, since notes generated against the master will refer to slides that are not in the cut. Reworking speaker notes you can read aloud is a per-cut job, not a once job.

And every cut has to carry the same visual identity, which is one more reason to derive rather than rebuild: the template work described in making an AI deck match the company template is done once on the master and inherited by all three.

The version problem

Three files, one topic, similar names. This is where multi-audience decks actually fail.

Four habits that cost almost nothing:

1. Name the audience in the file name, not v1, v2, final. "Q3 update, board" and "Q3 update, delivery team" are unambiguous under pressure.
2. Date every file, so the newest is identifiable without opening it.
3. Put the audience on the title slide of each cut. If the wrong file gets opened in the room, you find out in the first second rather than on slide nine.
4. Change the master first, always. If a number is wrong, fix it in the source, then reissue the cuts. Fixing it only in the version you happen to have open is how the discrepancy is created.

If you are only ever going to present one of the three live and send the others, consider whether the sent versions should be a document instead of a deck. Slides that have to be read alone are usually a worse artifact than a page of prose, and producing a page is faster than maintaining a third file.

Checks before each one goes out

• Do the figures match the master, exactly? Check the three most consequential in each cut, against the source, not against another cut.
• Is the ask present and early? Especially in the decision cut.
• Did any caveat disappear during shortening? Re-read anything a tool condensed, next to the original.
• Are the risks still there? Reordered is fine, removed is not.
• Does the opening slide name the audience and the purpose?
• Are the visuals and template consistent across cuts? A deck that looks different from its sibling suggests the content differs more than it does.

The underlying discipline is the same one any AI-built deck needs before it reaches an audience, described in building a work deck with AI. Multi-audience work just multiplies the surface, which is the argument for one source and three cuts rather than three decks that started life the same and drifted.

FAQ

Can AI make different versions of the same presentation for different audiences?
It can restructure, condense and retone an existing deck, which is most of the work of producing a cut. It cannot decide what your board needs to see or which risk is material to which audience, and it should not be trusted to preserve a caveat while it shortens. Use it for the shape and check the substance.

Is it better to make one deck that works for everyone?
Rarely. A deck built to satisfy three audiences is usually too long for the decision-makers and too shallow for the team. One master with three cuts costs less to maintain than one compromise deck costs in effectiveness.

How do I stop the versions drifting apart?
Keep the numbers in one source outside the slides, change the master before any cut, name files by audience rather than by version number, and regenerate charts from the data rather than copying images between decks.

Should the executive version just be the first five slides?
Not usually. The decision cut needs a purpose-written opening slide and a different order, not a truncation. Truncating the master gives you the background without the ask, which is the opposite of what that audience needs.

What about the detail I cut? Do I throw it away?
Move it to an appendix in the same file. It stays available for questions without being in the main sequence, and it means you are maintaining one body of content rather than deciding what to permanently delete.