Skip to content

THE POWERPOINT TEST · OCTOBER 2026

Six AI tools.
One consulting brief.
Every export, inspected.

Which AI presentation tool produces work a consultant can actually use? We asked for the same thirteen-slide deck, then inspected the storyline, sources, charts and PowerPoint exports.

By XLSlides EditorialRecorded October 2026
WATCH THE 9-MINUTE TESTOpen on YouTube ↗
THE SHORT ANSWER

In our recorded October 2026 consulting-deck test, XLSlides received the highest weighted score, Claude was the strongest of the two AI assistants, and ChatGPT stood out for research. The choice depends on the work: native chart editing, credible sources, storyline and cleanup time matter differently. We make XLSlides, and the test included important differences in settings and plans.

THE PUBLISHED SCORECARD

Our results, out of five.

Reviewer-assigned weighted scores from one recorded test. These are not independent ratings or statistically significant performance estimates.

01XLSlides3.95
02Claude3.80
03Perceptis3.625
04ChatGPT3.45
05Oria2.60
06Gamma2.00
ToolScore / 5What shaped the result
XLSlides3.95Annotated consulting exhibits and native charts; slowest run and illustrative data.
Claude3.80Strong editable output and charts; some headlines and visual polish needed work.
Perceptis3.625Consulting style and native charts; free-plan limit and an inconsistent forecast.
ChatGPT3.45Strong research and dated sources; dense output and repeated layouts.
Oria2.60One strong native slide; it did not complete the requested deck in this run.
Gamma2.00Fast browser presentation; the tested exported charts were pictures.
THE TEST RECORD
THE EXACT BRIEF“Make 13 slides on India’s quick commerce boom, covering key players and future outlook”

What this comparison actually tests

The question was whether an AI tool could turn a short brief into a consulting presentation that remains useful after export to PowerPoint. The brief requested thirteen slides about India’s quick-commerce market, its key players and future outlook. The same topic and target length made the outputs comparable, but the tools ran with their own interfaces, defaults and available plans.

This is one recorded benchmark, not a statistically controlled product evaluation. We did not repeat every run across several accounts, models or days. A result can be informative without establishing the average performance of a tool. The video was published on 18 October 2026; this article expands the record with screenshots and a reproducible review method.

We make XLSlides. We selected the criteria, assigned the scores and edited the comparison video. That commercial relationship is material to interpreting the ranking. Readers can inspect the recording and disagree with the assessment. The recording makes our reasoning reviewable, but does not turn an in-house evaluation into an independent review.

The conditions that affect the result

The video describes a single attempt for each tool and no uploaded reference template. Native themes and defaults were still different. Perceptis retained its built-in style; the PowerPoint assistants and browser tools used different workflows. These distinctions should remain visible rather than being smoothed into a claim that every condition was identical.

XLSlides ran with web search disabled. That is a substantial difference for a task involving market data. Its figures were illustrative, and the video calls out a wrong figure. The assistants’ ability to research cannot be compared directly with that configuration as though all six had been given the same information access.

Plan limits also affected completion. The recorded Perceptis free plan stopped at ten slides, and Oria generated one slide in the interaction shown. Those outcomes matter for the requested task, but they should not be presented as permanent limits on every version or plan. The exact plan/model identifiers are not fully documented in the available record, so this article does not invent them.

Finally, the published video description discloses that XLSlides slides 9 and 12 in the full-deck segment include later chart fixes. That segment is useful for inspecting the corrected output, while the disclosure prevents it from being mistaken for completely untouched first-pass output. Timing and cleanup claims should be assessed with that distinction in mind.

How the weighted scores work

Each criterion is scored out of five. The overall result is the sum of each criterion score multiplied by its weight. The seven weights total one hundred percent. Storyline, editability and avoidance of repetitive generated-looking output together carry sixty percent, which makes this a consultant-oriented judgment rather than a general design contest.

Speed carries five percent. That weighting explains how the slowest run could receive the highest overall score. Another reader may reasonably value speed more or view browser sharing as the main requirement. A different rubric could produce a different order; this article does not manufacture alternative scores without the underlying criterion-level data.

The values shown are the published weighted results. The difference between 3.95 and 3.80 is a reviewer score difference, not evidence of statistical significance. The labels concerning client readiness and visual repetition are subjective judgments. They should be read alongside the actual slides and the objective completion and export checks.

WHAT A CONSULTANT HAS TO FIX

The seven criteria.

CriterionWeightThe review question
Storyline & action titles20%Does the headline explain the implication rather than merely name the topic?
Data & sourcing10%Are figures sourced, dated and internally consistent?
Chart quality15%Does the exhibit use the right analytical form, labels and useful annotations?
PowerPoint editability20%Can a reviewer change the underlying data and edit the slide elements?
Free of AI slop20%Does the output avoid repetitive layouts, filler and unnecessary decoration?
Client readiness10%How much substantive and visual rework remains before a client review?
Speed5%How long did the recorded run take to produce its output?

THE RECORDED RESULT · 01

ChatGPT for PowerPoint

3.45 / 5
FRAME 1:54Recorded test, 1:54: ChatGPT’s self-review and output summary. The run used the selected high-capability model with extra-high reasoning; the record does not establish a complete model/version manifest.Image: XLSlides benchmark video, October 2026.

ChatGPT produced the requested 13-slide deck in the recorded run. The review highlighted dated source notes, research drawn from company filings and industry material, and native PowerPoint elements. It also showed an automated review pass addressing labels and text wrapping. Those are valuable behaviours for a team that needs evidence it can inspect.

The weaker point was the presentation’s narrative and visual economy. The reviewer described a research-report feel: dense text, repeated card arrangements and fairly simple bars and tables. A source-rich deck can still need substantial editing before its recommendation is clear. Native elements make that editing possible, but do not eliminate the work.

The video reports a run of about ten minutes. This is an observed duration for that run, not a service-level promise. Model choice, reasoning settings, network conditions and task complexity can change completion time. The exact model identifier is not fully documented in the material available for this article, so we do not substitute a guessed version.

Who should inspect this result closely: teams that value sourced first drafts and are prepared to refine the storyline and exhibits. Before using the deck externally, open each important source, check that it supports the particular number, and rewrite any headline that still describes a topic rather than a decision implication.

Inspect this moment in the recording ↗

THE RECORDED RESULT · 02

Claude in PowerPoint

3.80 / 5
FRAME 3:00Recorded test, 3:00: Claude’s slide output in the PowerPoint workflow. The article reports the tested Opus selection without inventing an exact version.Image: XLSlides benchmark video, October 2026.

Claude created 13 slides in the recorded PowerPoint workflow. The video shows a warm off-white theme, dark text and an orange accent, with a market-share chart, growth chart, competitor table and timeline. The reviewer counted ten action titles among the thirteen headlines and found the output sourced and editable.

Its score reflected a strong overall result among the two general-purpose assistants tested. The charts and native elements made it a credible starting point for a consultant who expects to keep working in PowerPoint. The remaining weaknesses were less about opening the file and more about finishing the message: a few headings remained labels, and some pages needed stronger visual hierarchy.

The recorded run took about twelve minutes. The video says an Opus model was selected and edits were allowed to proceed without repeated approval. Those conditions matter when interpreting speed and workflow. They do not establish that every Claude setup, plan or future version will behave the same way.

Who should inspect this result closely: consultants already working in PowerPoint who want editable charts and a coherent first draft. Test whether the headlines state the implication, whether sources match the displayed figures, and whether the chart style fits the client. A good native deck still needs a reviewer who owns the argument.

Inspect this moment in the recording ↗

THE RECORDED RESULT · 03

Gamma

2.00 / 5
FRAME 3:50Recorded test, 3:50: the selected Gamma chart appears under Picture Format. This is evidence about that exported object in this run, not a claim about every current Gamma export.Image: XLSlides benchmark video, October 2026.

Gamma’s browser presentation looked polished in the recorded comparison. It used a warm editorial theme, photography and some useful action-oriented headlines. The export itself was quick, and text remained editable when the file opened in PowerPoint. For a workflow ending in a browser presentation, those strengths can matter substantially.

The consultant-focused review penalised the exported charts. In the recorded check, selecting the market-size chart exposed picture formatting rather than a native chart data interface. The reviewer reported the same issue with another chart. If the audience asks for a changed assumption, an image cannot be updated by editing its embedded chart data.

That is a specific observation about the tested output. It should not be rewritten as a timeless statement that Gamma never supports editable content. Export behaviour can depend on the object, feature and version. The reproduction step is to inspect the actual exported chart in your own file and attempt the change you expect to make.

Who should inspect this result closely: readers deciding whether browser polish or native PowerPoint maintenance matters more. A browser-first team may accept a different trade-off from an analyst maintaining a recurring financial exhibit. The test’s low score reflects its consultant-oriented rubric, especially editability and analytical chart detail.

Inspect this moment in the recording ↗

THE RECORDED RESULT · 04

Oria

2.60 / 5
FRAME 4:40Recorded test, 4:40: Oria produced a native consulting-style slide. The main limitation in this run was completing one slide against a 13-slide request.Image: XLSlides benchmark video, October 2026.

Oria offered layout options and produced a slide with native shapes, logos and a clean three-column arrangement. The reviewer regarded the individual page as recognisably consulting-style. That is a different strength from generating a complete story: a single useful slide can be valuable in an established deck workflow.

The run did not complete the requested thirteen-slide presentation. Under this benchmark, that substantially limited client readiness and overall usefulness. A tool that solves one page well does not automatically solve a deck brief, even if its individual components are highly editable.

The result should remain scoped to the interaction shown. It does not prove that no other Oria workflow can produce more slides or that the product’s present capabilities are identical to the tested version. The relevant purchase test is to reproduce the intended end-to-end task, including how pages are added, how context carries forward and how the final deck is checked.

Who should inspect this result closely: PowerPoint users who need help with individual consulting exhibits and already own the deck narrative. Compare the time saved on a page with the work required to build the rest of the story. If the requirement is one prompt to a complete deck, completion should be a pass-or-fail check before subjective design scoring.

Inspect this moment in the recording ↗

THE RECORDED RESULT · 05

Perceptis

3.625 / 5
FRAME 5:45Recorded test, 5:45: the PowerPoint export exposes a native chart. The separate data-consistency concern remains even where the chart is editable.Image: XLSlides benchmark video, October 2026.

Perceptis delivered a consulting house style with action titles, KPI cards, a process diagram and takeaway bars. The reviewer described it as the most polished out of the box in this run. Its exported chart was native in PowerPoint, which is an important distinction from a screenshot of a chart.

Two limitations shaped the result. The tested free plan stopped at ten slides rather than the requested thirteen, and a later request encountered a paywall. The reviewer also identified a market forecast whose displayed values did not reconcile with its stated growth rate. A professional-looking chart remains unreliable if its arithmetic contradicts its own label.

Those are dated test conditions. The article does not treat the ten-slide limit as current pricing or a permanent product restriction. Readers should check the present plan and rerun the task they intend to purchase. Likewise, the observed forecast error is evidence about this generated deck, not a measurement of every output from the tool.

Who should inspect this result closely: teams that prefer a consulting visual style and want native charts, while retaining responsibility for calculations and source checks. Review both the full-deck limit and the economic logic of the exhibits. Editability helps correct an error; it does not establish that the data were correct in the first place.

Inspect this moment in the recording ↗

THE RECORDED RESULT · 06

XLSlides

3.95 / 5
FRAME 7:45Recorded test, 7:45: a native XLSlides chart is selected in PowerPoint. The video’s full-deck segment includes later chart fixes on slides 9 and 12, disclosed in its description.Image: XLSlides benchmark video, October 2026.

XLSlides received the highest weighted score in our review. The recorded deck used a range of analytical exhibits, including a unit-economics waterfall, a mix chart and scenario comparisons. The reviewer counted action titles throughout and showed editable titles, icons, tables and native PowerPoint charts.

The most important limitation was data. We ran XLSlides with web search disabled, while the assistants were allowed to research. Its numbers were therefore illustrative, and the video identifies a plainly wrong dark-store figure. The transcript’s automatic rendering of the company name is unreliable, so this article does not repeat that figure as a market fact. An illustrative label does not excuse an internally inconsistent or misleading claim.

It was also the slowest run, at about twenty-six minutes. The system performed visual checks and fixes, but the elapsed time is still part of the result. The published description additionally discloses that slides 9 and 12 in the full-deck segment use later chart fixes. They should not be presented as untouched first-pass evidence.

Who should inspect this result closely: consultants seeking annotated analytical exhibits and native editing, who will verify the evidence and can accept the recorded speed trade-off. We make XLSlides, chose the rubric and supplied the scores. The recording lets readers inspect our assessment; it does not make the assessment independent or remove the differences in test conditions.

Inspect this moment in the recording ↗

MAKE THE TEST YOURS

Choose for the work that happens after generation

For a consultant, the first draft is rarely the final deliverable. A client may change an assumption, ask for a different scenario or move one chart into another deck. That makes the cost of revision important. A visually polished browser page can be an excellent endpoint for one team and an awkward intermediate format for another.

If PowerPoint is the final working environment, test the exported file rather than relying on a vendor’s use of the word editable. Text, shapes, tables and charts can behave differently within the same deck. One editable title does not establish that every chart has an underlying data sheet. One flattened chart does not establish that every object is flattened.

If research is the bottleneck, inspect the sources and the argument first. A native chart filled with unsupported numbers is not ready for a client. If the storyline is the bottleneck, ask whether the titles form a coherent recommendation when read without the body text. If layout is the bottleneck, time the cleanup required on a representative page.

Our consulting presentation guide explains that decision-document standard. The executive summary tool and strategy recommendation builder offer narrower workflows when a full deck is not the task you need to solve.

A practical PowerPoint export check

Open the exported file in the version of PowerPoint your team actually uses. Select a chart and look for chart-specific controls, then try to edit a data value. Inspect a table cell, a title and an icon separately. Move an annotation and confirm that it still makes sense after the underlying figure changes.

Next, test a realistic revision. Replace a five-year forecast with a shorter horizon, add a competitor, or change the unit in an axis label. Watch for clipped labels, disconnected callouts and annotations whose meaning depends on the original number. A file can be technically editable while still being expensive to maintain.

Then save and reopen the revised presentation. Check that the edit persists and that the slide still renders correctly. Export a PDF if that is part of the delivery process. These steps test the whole handoff rather than only the moment when the download button completes.

CheckWhat to doWhat counts as evidence
Native chartChange a value through the chart’s data interfaceChart updates and remains editable
TableEdit an individual cell and resize a columnContent stays separate and legible
Source noteOpen the source and find the relevant figureCitation supports the displayed claim
LayoutApply a plausible client revisionLabels and annotations remain usable
HandoffSave, reopen and export the revised fileThe intended change survives

These checks are a proposed buyer workflow. They are not additional hands-on tests of all six products performed for this article. The screenshots above document the recorded comparison; current product behaviour should be verified in a fresh export.

Run a fairer test for your own team

Start with a brief that resembles the work you buy the tool to perform. If the team makes recurring financial reviews, use a small approved dataset and a defined decision. If it makes proposals, use a representative client situation with confidential details removed. A generic topic may not expose the limitations that matter in your workflow.

Record the date, plan, model identifier, settings, web access and any template or source material supplied. Set the same permissions and information access where possible. If a difference is inherent to the workflow, document it. Agree whether the task is one slide, a full deck or an update to an existing presentation.

Separate the first-pass result from the corrected result. Count the time spent waiting, reviewing, repairing sources and fixing the export. A tool that generates quickly can still have a high total preparation time. A slower tool may be acceptable if it removes more of the subsequent work, but that needs evidence from your task.

Use more than one attempt if the decision is material. Have a reviewer evaluate the output without seeing the product name where that is practical. Define pass-or-fail requirements before applying subjective scores. For example, completing the requested deck and preserving a critical editable chart may be mandatory even if the theme is attractive.

Read the market numbers as test material

The quick-commerce figures visible in these generated decks are not presented here as an authoritative market dataset. The benchmark itself identifies inconsistencies and illustrative inputs. A screenshot can faithfully document what a tool produced while still containing an incorrect number.

Before reusing a figure in client work, trace it to a dated primary source, confirm the market definition and check whether values are actuals, forecasts or scenarios. Distinguish gross merchandise value from revenue, and company-specific measures from market-wide totals. Recalculate growth rates rather than accepting an attractive chart label.

The same evidence discipline appears throughout The Consulting Files. The Swissair case examines a persuasive strategic map that needed a funding test; the Enron case asks why reported performance needs independent scrutiny. A generated slide deserves the same habit of asking what the exhibit actually proves.

What we would improve in the next benchmark

A stronger follow-up would record exact plan and model versions, use equivalent web access, repeat the runs and publish first-pass and corrected exports separately. It would measure total time to a verified client-ready deck, with source accuracy and editing tasks scored by more than one reviewer. It would also test a browser-first use case separately from a PowerPoint maintenance use case.

Those are proposed improvements, not work claimed to have been completed. The current article stays within the evidence of the October recording. Its value is a concrete task, visible output, explicit trade-offs and a test readers can adapt to their own needs.

Jump to the evidence.

THE FINAL TABLEThe scorecard in the recording at 8:22. All scores are the publisher’s own assessments under the disclosed rubric.Image: XLSlides benchmark video, October 2026.

SOURCES & PROVENANCE

What this article is based on.

  1. Original XLSlides comparison video and description ↗Published 18 October 2026. Source of the prompt, rubric, scores, stated conditions and disclosed chart fixes.
  2. Recorded rubric and test setup ↗The criteria and their weights; the tool segments show the output and the reviewer’s interpretation.
  3. Recorded final scorecard ↗The published weighted results. Screenshots on this page were captured from the recording and converted to WebP; the underlying product outputs were not recreated.

This article adds editorial interpretation and a proposed buyer checklist. It does not claim a fresh hands-on retest of all six products or complete preservation of their original exported files. Recheck current plans and behaviour before relying on the results for a purchase.

TRY THE SAME BRIEF

Inspect the output for yourself.

Use the original brief, verify every market figure and test the PowerPoint export against your own workflow. Judge the result by the work that remains.

Run the brief in XLSlides

Check current XLSlides plans · Read the focused Gamma comparison