AI PRESENTATIONS · OCTOBER 2026 · REPORT ISL-20261006-001

SlidesGPT review:
Company, pricing & AI testing
A six-slide business brief, supplied financial facts, and ten checks fixed before generation.
Page 2 · Full company & testing review
Company and product profile
SlidesGPT offers browser-based AI presentation generation. Its imprint identifies Lightstone GmbH in Cologne, Germany, and lists Benjamin Spinola as shareholder. These are vendor-published details, not an independently audited ownership record. Company imprint.
The About page says the company was founded in 2023 and reports more than ten million presentations. We have not independently verified that usage figure or its customer-logo claims. About SlidesGPT.
Pricing and exports
The pricing page lists free creation, viewing and sharing. Its displayed annual Pro option is $89.99 per year ($7.49/month equivalent), with 10 downloads per month and PowerPoint, PDF and Google Slides exports. These are vendor-listed terms checked October 6, 2026; checkout, taxes, billing changes and export fidelity were not tested. Current pricing.
Privacy documentation
The privacy page describes device and usage data collection and states that conversation and personal data may be kept for up to six years unless an account is deleted. The imprint identifies a German operator while the privacy page describes the country as United States. We have recorded this documentation inconsistency for clarification; it is not a security finding or a compliance conclusion. No private company material was submitted in this pilot. Privacy policy.
Test design and grading
This is a controlled fact-preservation and instruction-following task in the AI presentations category. The fictional dataset makes expected answers inspectable without live research.
- Each of ten criteria earns 10 points only if fully met; otherwise 0.
- Structure: C01, C02, C07 (30 points). Facts and arithmetic: C03–C06, C09 (50 points). Caveats: C08, C10 (20 points).
- “On slide 3,” “on slide 4” and “on slide 6” mean actual rendered slide positions. Speaker notes do not count as visible slide content for those requirements.
- Grade the first output without editing it. Preserve failed criteria and the evidence supporting them.
- Missing evidence stays unscored. This run had all ten checks observable.
The category grouping and clarifications above were documented at review time; the ten criteria and point rules were saved before submission. Any retest must declare its configuration in advance.
Exact submitted prompt
Output evidence · all eight rendered slides
Text captured from the live browser-rendered deck. Speaker notes are identified in each excerpt. The full raw capture is available in the results JSON.
Slide 1 · Cedar Analytics
Slide 2 · Agenda
Slide 3 · Cedar Analytics
Slide 4 · Product and Pricing
Slide 5 · January vs February
Slide 6 · February Operating Snapshot
Slide 7 · March Target
Slide 8 · Limitations and Next Steps
PROPOSED BENCHMARK · NOT EXECUTED
Extensive SlidesGPT testing plan
This is the long-form testing program: 60 checks across 12 areas. Every item below is Not tested under this new benchmark. The existing ten-check pilot remains separate. Company website information supplies background and disclosed terms; it cannot substitute for measured product results.
How the full rating will work
- Lock input fixtures, answer keys, criterion-specific 0–4 rubrics, settings, tiers and category weights before execution.
- Use three task variants and three fresh runs per variant for generation-dependent checks. Retain every output, including failures.
- Use two blinded reviewers for subjective checks; record disagreements and their resolution.
- Proposed category points equal the category weight multiplied by the sum of its five check scores divided by 20. Weights total 100 points.
- Not tested does not mean failed or zero. Record unsupported features and fix applicability rules before matched comparison runs. Keep paid-tier results explicitly labeled.
- Withhold the overall product score until required coverage is complete. Any certification threshold must be published and supported by evidence before an award is issued.
Instruction following
5 planned checks · Proposed weight 10/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| INSTRUCTIONS-01Exact slide-count control Method: Generate the same six-slide brief using the explicit count setting and Auto separately, three fresh runs each. Evidence: Rendered count, cover/agenda behavior, settings and every output. | NOT TESTED |
| INSTRUCTIONS-02Required headings and order Method: Provide six exact headings; compare every rendered title and position with the fixture. Evidence: Ordered title list and position-match percentage. | NOT TESTED |
| INSTRUCTIONS-03Required content placement Method: Request a table, chart and limitations on specified slides; count covers and agenda consistently. Evidence: Rendered positions and screenshots of each required element. | NOT TESTED |
| INSTRUCTIONS-04Formatting constraints Method: Request a fixed bullet count, word limit and prohibited phrases on designated slides. Evidence: Bullet counts, word counts and prohibited-phrase occurrences. | NOT TESTED |
| INSTRUCTIONS-05Conflicting instructions Method: Supply a deliberately conflicting brief and a clear priority rule; observe whether the tool follows the rule or requests clarification. Evidence: Conflict resolution and unsupported decisions, with original prompt. | NOT TESTED |
Factual accuracy
5 planned checks · Proposed weight 15/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| FACTS-01Source-fact preservation Method: Supply a fictional brief with 20 locked facts, names, dates and amounts; compare output with an answer key. Evidence: Correct facts out of 20, omissions and alterations. | NOT TESTED |
| FACTS-02Unsupported claims Method: Use a brief with intentionally limited information and prohibit invented facts. Evidence: Every unsupported factual claim and its slide location. | NOT TESTED |
| FACTS-03Missing-information handling Method: Omit figures needed for a conclusion and explicitly prohibit estimation. Evidence: Whether gaps are disclosed, questions are asked or numbers invented. | NOT TESTED |
| FACTS-04Citation and link validity Method: Provide a set of controlled source URLs and require source attribution; inspect cited links and supporting passages. Evidence: Working links, supported claims and fabricated citations. | NOT TESTED |
| FACTS-05PDF extraction accuracy Method: Upload an authorized synthetic PDF containing known paragraphs, tables and footnotes, when the tested tier supports it. Evidence: Extracted facts versus fixture; missing or distorted table/footnote content. | NOT TESTED |
Calculations and data integrity
5 planned checks · Proposed weight 10/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| MATH-01Growth calculations Method: Use locked positive, negative and zero-baseline examples with known answers. Evidence: Correct calculations, units, rounding and zero-baseline handling. | NOT TESTED |
| MATH-02Totals and percentages Method: Provide a synthetic dataset with totals and percentage breakdowns that must reconcile. Evidence: Absolute errors and whether percentages/totals reconcile. | NOT TESTED |
| MATH-03Money and unit consistency Method: Mix monthly/annual costs, two currencies and thousands/millions with explicit conversion rules. Evidence: Unit labels, currency labels and conversion accuracy. | NOT TESTED |
| MATH-04Chart-to-table agreement Method: Require a chart and table from the identical dataset. Evidence: Every chart label/value compared with the source table. | NOT TESTED |
| MATH-05Targets versus actual results Method: Mix achieved results, forecasts and goals, explicitly labeled in the fixture. Evidence: Misclassification count and visibility of forecast/target labels. | NOT TESTED |
Writing and communication
5 planned checks · Proposed weight 8/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| WRITING-01Audience adaptation Method: Generate the same brief for executives and beginners; evaluate against a rubric fixed before generation. Evidence: Blind rubric scores for terminology, explanation and audience fit. | NOT TESTED |
| WRITING-02Narrative flow Method: Use a problem-evidence-options-recommendation brief and inspect the causal sequence. Evidence: Sequence adherence, transitions and unexplained conclusions. | NOT TESTED |
| WRITING-03Conciseness and readability Method: Specify slide word limits and readability requirements on a fixed source brief. Evidence: Words per slide, repeated content and predefined readability checks. | NOT TESTED |
| WRITING-04Technical explanation Method: Provide a technical passage with an answer key and ask for an accessible explanation. Evidence: Key concepts retained and meaning-changing simplifications. | NOT TESTED |
| WRITING-05Multilingual fidelity Method: Translate the same locked brief into specified languages and use qualified reviewers. Evidence: Name/number preservation and reviewer-graded meaning, fluency and terminology. | NOT TESTED |
Visual design and layout
5 planned checks · Proposed weight 8/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| DESIGN-01Overflow and clipping Method: Generate short, medium and long text fixtures; inspect at fixed display dimensions. Evidence: Clipped text, collisions and elements outside slide bounds. | NOT TESTED |
| DESIGN-02Typography and hierarchy Method: Assess title/body/caption styling using a predeclared readability rubric. Evidence: Font sizes, hierarchy consistency and unreadable labels. | NOT TESTED |
| DESIGN-03Theme consistency Method: Generate multiple slides with the same chosen theme and review spacing, color and alignment. Evidence: Style deviations across the deck, supported by screenshots. | NOT TESTED |
| DESIGN-04Brand adherence Method: When supported, supply a test logo, palette and fonts owned by iSync. Evidence: Exact palette/font/logo placement checks and unsupported controls. | NOT TESTED |
| DESIGN-05Image relevance Method: Use five synthetic scenarios and grade supplied/generated imagery against the brief. Evidence: Relevant images, misleading imagery and documented provenance where available. | NOT TESTED |
Charts, tables and diagrams
5 planned checks · Proposed weight 7/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| CHARTS-01Table completeness Method: Request a table with fixed rows, columns, labels and values. Evidence: Missing cells, changed values and visual legibility. | NOT TESTED |
| CHARTS-02Chart suitability Method: Use time-series, categorical and proportion datasets with permitted chart types defined in advance. Evidence: Chart-type match and whether the chosen representation misleads. | NOT TESTED |
| CHARTS-03Axes and labels Method: Require units, labels, scales and source notes for each chart. Evidence: Correct axes, omitted units and misleading truncation. | NOT TESTED |
| CHARTS-04Dense data readability Method: Supply a controlled larger table and inspect both displayed and exported slides. Evidence: Readable values, overlap and handling of overflow. | NOT TESTED |
| CHARTS-05Diagram relationship accuracy Method: Request a flowchart from a locked process with branches and exceptions. Evidence: Correct nodes, edges, branch labels and omitted exceptions. | NOT TESTED |
Editing and user control
5 planned checks · Proposed weight 7/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| EDITING-01Outline editing Method: Change a heading and reorder a section before generating the deck. Evidence: Whether final output preserves the saved edits and order. | NOT TESTED |
| EDITING-02Slide-level revision Method: Request a change to one slide while explicitly preserving all others. Evidence: Target change completed and unintended changes elsewhere. | NOT TESTED |
| EDITING-03Regeneration stability Method: Regenerate a selected element with facts and numbers declared invariant. Evidence: Preserved facts, retained layout and unsupported changes. | NOT TESTED |
| EDITING-04Undo and recovery Method: When offered, make and undo a sequence of edits on a disposable test deck. Evidence: Recoverable states, restored content and lost edits. | NOT TESTED |
| EDITING-05Save and reopen Method: Save an authorized test deck, close the view and reopen it through the product UI. Evidence: Content, styling and notes retained after reopening. | NOT TESTED |
Exports and compatibility
5 planned checks · Proposed weight 8/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| EXPORTS-01PowerPoint fidelity Method: Export an authorized test deck to PPTX on a tier that permits it; inspect it in PowerPoint or another declared viewer. Evidence: Text, layout, charts, notes and editability versus browser output. | NOT TESTED |
| EXPORTS-02PDF fidelity Method: Export the same deck to PDF and inspect every page. Evidence: Page count, clipped elements, fonts and text selection. | NOT TESTED |
| EXPORTS-03Google Slides fidelity Method: If available, import/export the test deck through an authorized Google Slides account. Evidence: Layout and editability differences from the source deck. | NOT TESTED |
| EXPORTS-04Font and media portability Method: Open exported files on a second declared environment without the original fonts installed. Evidence: Fallback fonts, missing images and broken media. | NOT TESTED |
| EXPORTS-05Notes and link preservation Method: Include known speaker notes and links, then inspect each supported export. Evidence: Notes retained, links functional and content in the correct layer. | NOT TESTED |
Speed and repeatability
5 planned checks · Proposed weight 7/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| RELIABILITY-01Generation time Method: Measure from submission to usable completed deck across repeated locked tasks. Evidence: Instrumented timing, median, range and environment/tier. | NOT TESTED |
| RELIABILITY-02Fresh-run consistency Method: Run three predefined task variants three times each without selecting the best output. Evidence: All nine results and variation in checklist scores. | NOT TESTED |
| RELIABILITY-03Failure and retry behavior Method: Record normal failures during the planned runs and exercise documented retry controls. Evidence: Failure rate, error messages, recovery and duplicate outputs. | NOT TESTED |
| RELIABILITY-04Long-brief handling Method: Use increasingly long synthetic briefs within documented product limits. Evidence: Completion, omissions, truncation and time by input length. | NOT TESTED |
| RELIABILITY-05Version-aware retesting Method: Repeat the locked suite after a documented product update or at the scheduled retest date. Evidence: Configuration/version evidence and comparable historical outputs. | NOT TESTED |
Usability and workflow
5 planned checks · Proposed weight 4/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| USABILITY-01First-run completion Method: Have a first-time tester complete a declared task without assistance. Evidence: Time, errors, help required and completed steps. | NOT TESTED |
| USABILITY-02Settings discoverability Method: Ask testers to locate slide count, theme, language and export controls that exist on the tier. Evidence: Task completion and navigation steps. | NOT TESTED |
| USABILITY-03Mobile workflow Method: Generate, inspect and edit where supported on a declared mobile viewport/device. Evidence: Usable controls, blocked steps and display defects. | NOT TESTED |
| USABILITY-04Plan-limit clarity Method: Compare displayed generation/download limits with actual behavior on the authorized tier. Evidence: Clear limits, blocked actions and unexpected restrictions. | NOT TESTED |
| USABILITY-05Support response Method: With owner authorization, submit one ordinary support question about an observed issue. Evidence: Response time, relevance and whether the proposed fix works. | NOT TESTED |
Accessibility
5 planned checks · Proposed weight 4/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| ACCESSIBILITY-01Keyboard navigation Method: Complete the declared core workflow using only the keyboard. Evidence: Reachable controls, focus visibility, traps and completion. | NOT TESTED |
| ACCESSIBILITY-02Screen-reader labels Method: Inspect the core UI and generated document with a named screen reader. Evidence: Meaningful control names, reading order and missing labels. | NOT TESTED |
| ACCESSIBILITY-03Text contrast Method: Measure text/background combinations on generated slides against a declared contrast threshold. Evidence: Measured ratios and failures by element. | NOT TESTED |
| ACCESSIBILITY-04Reading order and alternatives Method: Inspect exported reading order and alternative text where supported. Evidence: Logical reading sequence and informative image alternatives. | NOT TESTED |
| ACCESSIBILITY-05Zoom and small-screen legibility Method: Inspect the UI at 200% zoom and the deck on a small screen. Evidence: Lost controls, overlap and unreadable content. | NOT TESTED |
Privacy controls and documentation
5 planned checks · Proposed weight 2/100 · No category rating yet
| Check, method & required evidence | Status |
|---|---|
| PRIVACY-01Disclosure consistency Method: Compare the operator identity, privacy terms and in-product data notices. Evidence: Contradictions or unclear statements, labeled as documentation findings. | NOT TESTED |
| PRIVACY-02Sharing-default visibility Method: Create a synthetic deck and inspect default sharing controls and public/private labels. Evidence: Who can access the deck under each available setting. | NOT TESTED |
| PRIVACY-03Deletion-control behavior Method: Use only an explicitly disposable synthetic deck and the documented deletion workflow. Evidence: UI confirmation and subsequent accessibility under the same account. | NOT TESTED |
| PRIVACY-04Retention and training options Method: Inspect available data-retention and model-training preferences without assuming vendor claims are verified. Evidence: Displayed choices, changes applied and remaining uncertainty. | NOT TESTED |
| PRIVACY-05Sensitive-data guidance Method: Inspect warnings and instructions about submitting private material; submit only synthetic fixtures. Evidence: Clear user guidance and observable controls; no security-assurance conclusion. | NOT TESTED |