AI tools have exploded onto the scene and there are now a dime-a-dozen that can generate presentations on any topic in seconds. The tools have evolved quickly and many of them are genuinely impressive and useful for a first scaffold (see my own test of some tools in our previous article).
But a deck that looks finished is not the same as a deck that is precise, outcome-driven, and built around a story that holds together. The gap between an AI-generated deck and a top-tier consulting deck is consistent and predictable, and once you can name it, you can fix it in your own work.
In this article, we’ll first go over what an AI-generated deck is and why it falls short of consulting-grade decks. Second, perhaps more useful, we’ll show you how to actually use AI well when creating presentations so you can create McKinsey-level decks faster and better. And a caveat throughout; our focus is on consulting-type presentations (think data-heavy, structured, report-like) and how to make these match the level you’d expect from top-tier consultants.
An AI-generated deck is a presentation produced by a large language model from a short prompt with or without a data dump, where the tool drafts the structure, the slide text, and often the layout in one pass. The model predicts the most likely next words based on the thousands or millions of presentations it was trained on.
That prediction mechanism is the root of every issue below. A model optimizes for the statistical average of its training data, so its default output is, by construction, a generic version of a deck and generated anew every time.
This is worth stating plainly because it reframes the problem. The weakness of an AI deck is not a bug to be patched in the next release; it is a direct consequence of how the tool works. Knowing that tells you exactly where your own judgment and process has to take over.
After testing a bunch of AI slide generators and researching what our peers have tested, we’ve found seven main ways that AI-generated decks differ from human-made top-tier consulting decks from the likes of BCG, Bain, and McKinsey.
An AI model generates the most statistically likely content, which is by definition the most generic. This means a lot of the output can feel cookie cutter and lack the small, but crucial layout changes that suit your particular situation.
AI tools tend to just jump right into slide creation, without considering the wider strategic picture. In a consulting project, you’d do it the other way around, starting with a day-one-answer and hypothesis, building slides and filling in data to prove or disprove that, and finally rejigging everything to be executive-ready.
AI decks can suffer from “sudden jumps in logic” or “three slides that essentially say the same thing in slightly different words,” because the model lacks slide-to-slide continuity and overall storyline rigor. In a consulting deck, each slide’s action title links to the next, so the titles alone read as one continuous argument.
In addition, the job of a consultant is to prune and rebuild the storyline over and over so it continues to build toward the desired outcome in the best possible way as new data is analyzed, the deck is pressure-tested with key stakeholders, and the arguments are practiced out loud to gauge continuity and impact. AI-generated decks tend to add more slides, not systematically prune, test, and redo.
Text overload and general visual clutter is the most visible sign of an unedited AI deck, because the model defaults to thoroughness and packs each slide with bullet points and graphics. A consulting slide in contrast makes one point and moves the supporting details to the appendix. The white space in a consulting slide is just as important as the filled space, precisely because it helps emphasize the message of the slide.
AI slide generators tend to overfill slides leading to the main takeaway being harder to see upfront.
AI can fabricate statistics and state them so confidently that it can be both confusing and lead to time-consuming editing afterward. In particular, the anchoring-and-adjustment effect where you tend to iterate on an already stated phrase or fact instead of thinking about it from scratch leads to a lot of second-guessing on what to include in the slides and time wasted searching for data that isn’t important for the so-what or conclusion.
AI copy is “notoriously wordy” and leans on filler like “in today’s fast-paced world,” sounding sophisticated while missing the specific constraints of the actual problem. Nobody loves lengthy texts like LLMs! Consulting writing is the inverse: precise, to the point, and specific to the situation. The skill of a great consultant is the editing down to as little text as possible that still gets the main takeaway or so-what across.
And finally, AI slide generators (at least the ones I’ve tested) are great at an 80% version but miss the final crucial details that actually make it a consulting-grade slide. These are things like the amount of white space, the no-fly-zone, the balance between font sizes, which elements and conclusions are highlighted etc. All these small details may seem insignificant on their own, but they compound and make the overall result feel “off”.
AI slide generators miss the last 20% formatting that makes each slide feel consulting-grade.
In my experience at McKinsey, there were six main steps to creating a stellar presentation that achieved the desired results:
Decide who the deck is for and what you want it to lead to; a go-decision, a green light, a shared view of the current situation, a budget approval. Everything downstream serves this.
Build the section-level structure to fit the audience type and geared to achieve the outcome you defined. This is the skeleton the whole deck hangs on.
Sketch the individual slides within each section, including a first take on action titles and a rough sense of what data each slide will need or where to find it.
Build each slide, insert the data, write the text, and iterate and analyze as you go. This step is usually the largest time sink in the whole process.
Cut anything that does not move the story forward. Check that every action title fits the overall structure, and remove slides that repeat a message. This is one of the most overlooked skills of a good consultant.
Spacing, alignment, font sizes, consistent icons, spelling, missing footnotes. They seem minor, but together they create the impression that you are detail-oriented and have left no stone unturned.
Most AI presentation tools tackle steps 2-4 in one go. They ask you to prompt or upload data and then generate the whole structure, slides, and text in a single (sometimes long) sitting. Others focus narrowly on step 4, prompting to auto-fix a single slide, and/or on step 6 to check the final deck.
Collapsing steps 2 to 4 into one prompt is the heart of the problem for several reasons. First, it removes the crucial critical thinking where you actively construct a storyline and arguments from first principles and the exact situation you’re in. Second and related to critical thinking, most AI slide generators overfill and hallucinate content that forces you into an anchoring-and-adjustment pattern and makes it much harder to build a deck with only the absolute crucial data in it. Third, the tendency of AI slide generators to produce wordy, fluff-filled, cluttered slides with small details that are “off” means the clean-up time is surprisingly long and you end up spending almost as much time editing the slides as you would have building them from scratch.
The result is the generic, overfilled, loosely sequenced deck described earlier that looks good at first glance but turns out to be largely unusable when you’re actually sitting and building the final slides. Not because the tool is bad, but because it is pointed at the wrong part of the workflow.
With all that said it sounds like we are anti-AI. We are definitely not! Used in the right way, AI can be a magical assistant that speeds up the mechanical parts of deck-building without leaving its fingerprints on the thinking. Like having a second-year, super-speed analyst at your beck and call at any moment. Here is how we use it as consultants ourselves:
A model cannot decide who you are presenting to or what you need from them. Answer two questions before you open any tool: who is the audience, and what should the deck get them to do. Use AI only to pressure-test your answer, not to supply it.
And remember the small human differences like considering if the audience can deliver that outcome / answer you need, what objections or concerns the audience might have, and what evidence can counter those objections and concerns.
Ask a model to propose a few section-level structures for your audience and outcome, or come up with a couple of logical flows that will lead to your desired outcome (e.g., give me an SCR storyline and a pyramid principle storyline version of “x”). Treat the output as options or a first draft. Then keep going back to the overall storyline and prune, shuffle, and redo slides as you gather data. And always, always, always test it with select stakeholders or key people around you so you can get a feel for how it plays out in real life.
AI is useful for proposing what each slide could show and drafting a first take on action titles. This gives you a working scaffold to react to, which is far faster than starting each slide from blank.
Step 4 is the biggest time sink, so it is where good AI assistance pays off most. The right help here is not “generate the whole deck” but “suggest a strong way to show this specific point” and “insert a first pass of the text and data so your job becomes adjusting, adding, and thinking.” Again, think of it as an analyst you can direct as you wish.
And remember to always find the data yourself and sanity check everything!
Have a model check whether each action title fits the overall storyline and flag where it does not. Ask it to surface slides that carry roughly the same message so you can merge or cut them. You make the final call on what goes; AI just makes the candidates visible.
This is the step AI handles most reliably. Use it to catch double spaces, inconsistent fonts and alignment, spelling and grammar, and missing footnotes or data. It is genuinely good at the detail pass that humans rush at the end.
The principle behind all six steps is to treat AI as an assistant and slide monkey that you direct and iterate with. It can assist across the entire workflow but don’t try to compress steps 2-4 into one single prompt and expect a full, finished, McKinsey-level deck. In other words, keep the thinking yourself and delegate the mechanics.
AI tools can support your slide workflow effectively if you use them correctly.
A consulting deck is precise, sourced, and built around a single outcome-driven storyline. An AI-generated deck tends toward generic content, overfilled slides, hallucinated statistics, and breaks in narrative because it predicts the average rather than reasoning toward a conclusion.
The fix is not to avoid AI but to point it at the right steps: keep steps 1, 2, and 5 firmly in your own hands, and lean on AI for the slide-building and hygiene work in steps 3, 4, and 6.
In other words, think of AI slide tools as first-year consultants: quick and willing to do the grunt work, but over-eager and missing key learned skills like storyline rigor, sanity checks, less-is-more and the ability to understand the audience properly. Put yourself in the position of senior person who can pressure-test and sanity check the output and AI tools can 5-10x your productivity without losing the quality.
Most AI tools try to generate slides from scratch. Slideworks takes a different approach: instead of generating slides, you pull from a library of full-length, MBB-grade layouts built by ex-McKinsey, BCG, and Bain consultants so step 4, the biggest time sink, starts from best practice rather than a blank page.
Get your own library today
Q: Can AI make a good consulting deck?
A: AI can draft a structure and first-pass text quickly, but it cannot supply the diagnosis, sourcing, and storyline a consulting deck requires, so it works best as an assistant rather than the author.
Q: What is the right way to use AI to make slides?
A: The right way is to keep the audience definition, storyline, and editing in your own hands, and use AI for drafting slide stubs and action titles, suggesting layouts, inserting first-pass text and data you verify, and running final hygiene checks.
Q: Why do AI-generated slides look generic?
A: AI-generated slides look generic because the model produces the most statistically likely content from its training data, which is by definition the average rather than a specific, differentiated point of view.
Q: How do you tell an AI-generated deck from a consulting deck?
A: An AI-generated deck typically has overfilled bullet slides, topic-label titles, unsourced statistics, and logic jumps between slides, while a consulting deck has single-message slides, full-sentence action titles, and a sourced figure on every data slide. In other words, a consulting deck feels crisper and tighter than an AI-generated deck.
Q: Which presentation steps should humans keep from AI?
A: Humans should keep three steps: defining the audience and outcome, structuring the storyline, considering the analysis and so-whats, and editing ruthlessly; AI is best used for drafting slides, suggesting layouts and titles, and running hygiene checks.
