remio vs Manus vs GenSpark: a head-to-head on McKinsey-style AI presentation generation

TL;DR: We gave Manus, GenSpark, and remio the exact same 12-page deck script written in McKinsey editorial style — same titles, same bullet text, same one-line layout suggestion per page, and an explicit instruction “all text must come verbatim from the script.” GenSpark ignored the verbatim rule and fabricated extra content on every page. Manus respected the script but used a Material-Design web palette that drifts from McKinsey editorial code. remio matched both the script and the editorial look. Side-by-side images below.

Why McKinsey style still matters in the AI deck era

McKinsey, BCG, and Bain decks look the way they do for a reason. Single accent color. Hairline separators. Generous whitespace. No decorative gradients or AI-looking icons. Action-titled headlines that state a judgment, not a topic. Structured frameworks — the 2×2 matrix, the temple, the waterfall — rendered with their visual conventions intact. This is information density without visual noise: a board director reads a McKinsey slide in 8 seconds and walks away with one judgment.

When most AI agents generate “presentation slides”, what you get instead is a Material-Design-flavored web layout — gradients, soft drop shadows, candy-bright status colors, decorative icons, three-card layouts that all look the same. It looks fine on a portfolio site. It looks wrong in a steering committee.

This post is a head-to-head benchmark of three AI agents on a single brief: produce a 12-page McKinsey-style consulting deck on “Rebuilding consulting services in the AI era.” Same prompt, same script, same length, no per-slide hand fixes.

The setup

• Prompt: an explicit 12-page script (P1 cover, P2 agenda, P3 executive summary, P4 big number, P5 comparison table, P6 before/after, P7 four columns, P8 temple framework, P9 vertical steps, P10 timeline, P11 key takeaway, P12 next steps meeting agenda), with every page’s title, fields, and a one-line layout suggestion (e.g. “Spacious cover”, “Temple framework (roof + pillars + foundation)”). The prompt explicitly said “all text must come verbatim from the script.”
• Style brief: McKinsey editorial — restrained typography, navy primary, no decorative gradients, hairline separators, whitespace.
• Agents tested: Manus, GenSpark, remio.
• Output: PPTX → PNG renders. No retries, no manual edits.

Finding 1 — GenSpark misses on both content fidelity and visual style

GenSpark fails on two independent dimensions at once: it rewrites the brief on every page, and it renders everything in a SaaS / Material-Design visual vocabulary that has nothing to do with McKinsey editorial code. It is not a McKinsey-style deck generator at all — it is a generic “AI presentation” generator that branded its output with a fake “McKinsey & Company” logo.

P1 cover — GenSpark added an entire dashboard the script did not request

GenSpark

The script said “Title / Subtitle / Author·date”. GenSpark added: a fake “McKinsey & Company” logo top-left, a “Strategy Research” header top-right, a baby-blue “AI Transformation Series” tag pill, four invented stat cards (“68% AI output share”, “24% annual growth”, “12 core capabilities”, “3yr roadmap”), a “Confidential — For Internal Use Only” footer, a page-number “Page 1 of 12”. None of these were in the brief. The cover is supposed to set a single focal point; GenSpark spoiled the deck’s three biggest numbers before page 2.

For comparison, remio’s cover on the same brief keeps only the three requested fields — title, subtitle, attribution — left-aligned with generous whitespace:

remio (correct)

P2 agenda — invented duration tags and category pills

GenSpark

The script gave five chapters with names and one-line descriptions. GenSpark added: fake duration labels (“15 min / 20 min / 25 min / 20 min / 15 min”) that no one asked for, two color-coded category pills under each row (“Strategic Analysis”, “Market Trends”, “Client Insights”…), background circular decorations, and the persistent fake “McKinsey & Company” header. The brief never mentioned this is a meeting, never mentioned durations, never asked for sub-tags.

remio’s agenda page on the same brief is just numbered rows with the chapter name and description — no fabricated metadata:

remio (correct)

P3 executive summary — turned a list into a “feature grid”

GenSpark

The script asked for one emphasized headline banner over a numbered list of four takeaways. GenSpark turned the four takeaways into four feature cards with green/blue/purple/orange status pills (“68%”, “2x”, “Active”, “3yr”) in the top-right corner of each card. The judgment-list-under-banner structure that McKinsey executive summaries are famous for is gone — it now looks like a SaaS landing page.

remio renders the same content as a navy headline banner with the four takeaways stacked vertically below — exactly the executive summary pattern:

remio (correct)

And the visual style is a separate problem

Even setting aside the off-script content, the look of every GenSpark page is wrong for McKinsey editorial style:

• Soft gradient backgrounds instead of plain white whitespace. McKinsey decks use the white of the page as a structural element; GenSpark fills it with a baby-blue-to-white wash on every slide.
• Pastel status pills in card corners. Green, blue, purple, orange tags floating in the top-right of every card. McKinsey decks have at most one accent color per page, never four.
• Material-style rounded card containers with drop shadows. Each piece of content is wrapped in its own white rounded rectangle with a subtle shadow — a Material Design 3 / Stripe / Notion convention. McKinsey decks use hairline separators and surface fills, never rounded shadow cards.
• Decorative icons in colored circles. Tiny icons next to each item in green/blue/purple/orange backgrounds. McKinsey uses icons sparingly and never as a category-color carrier.
• A fake “McKinsey & Company” header and footer. A square “M” logo top-left and “Confidential — For Internal Use Only / Page X of 12” footer on every page. The prompt never asked for this. It is impersonation, not styling.
• Bright dashboard signal colors (#10B981 green / #F97316 orange / #A855F7 purple / #3B82F6 blue) used as a categorical palette. McKinsey palettes are deep, low-saturation editorial hues — deep navy, deep forest green, deep ochre, deep brick red.

The combination — script overridden + visual code that says “SaaS dashboard” — means GenSpark is currently the wrong tool when the requirement is “McKinsey-style.” The output is competent design in a different aesthetic, not a faithful McKinsey deck.

Finding 2 — Manus follows the script but uses the wrong color and layout vocabulary

Manus is the better-behaved of the two competitors: it respects the verbatim rule. The text content on Manus’s pages matches the script almost exactly. But the visual language is not McKinsey editorial — it is generic Material-Design web style with one blue accent.

P4 big number — Manus’s “68%” looks like a marketing site, not a consulting page

McKinsey reference visual code (remio)

Manus

The differences:

• Filled accent tile vs floating digits. remio puts “68%” inside a deep navy filled rectangle — the standard McKinsey “big-number tile” treatment. Manus uses a thin sans-serif “68” with a separate blue “%” floating to the right; no containing tile, the focal weight is weak.
• Accent color. McKinsey navy is #051C2C (or sometimes #1F4E79) — saturated but heavy, not "tech blue." Manus's accent is a bright Material blue (#2563EB-ish), which appears on every page (small dashes above titles, the "%" symbol, dots in front of bullets). Bright blue is the default Material/Stripe palette — it reads as a SaaS product page, not as a consulting analysis.
• Decorative rules. Manus adds a black hairline above the title and another at the very bottom of the slide; in between, short blue dashes float around the layout. These dashes — top-right above the title block, in front of every bullet — are pure decoration with no informational role. McKinsey decks deliberately do not do this.

P12 next steps — Manus’s table reads as a tracker, not a meeting agenda

remio

Manus

The brief asked for a meeting agenda table with row states (key, normal, break) — McKinsey's agenda treatment is a tabular list with subtle row-state styling (a left accent bar for "key" rows, a surface-tinted band for "break" rows, a small pill at the right for the status word).

remio rendered exactly that: navy left-bar on W1/W4/W6 (the key rows), a soft surface-gray band on W3 (the break row), and a small navy “key” pill on the right of each key row.

Manus rendered a generic data table — blue weeks, identical row formatting top to bottom, blue/grey/dark-blue underline indicators next to the status word. Functionally correct, but the “key vs normal vs break” hierarchy that the meeting-agenda treatment is designed to communicate is invisible at a glance.

P10 timeline — Manus’s milestones zig-zag above and below the axis without reason

remio

Manus

Both render a horizontal axis with five numbered dots. remio keeps every milestone above its dot, in the same row, with the label beneath — clean linear left-to-right read. Manus alternates above/below — 2025 Q4 above, 2026 Q2 below, 2026 Q4 above, 2027 Q2 below — which is a stylistic choice but turns a clean timeline into a reading zig-zag. McKinsey timelines stay on one row; Manus’s choice prioritizes “visual interest” over information clarity.

Finding 3 — remio’s edge comes from a curated McKinsey layout library, not from a better LLM

All three agents are calling foundation-model LLMs of similar capability. The gap is not in model quality. It is in how each agent represents a slide internally:

• Manus and GenSpark treat each slide as a free-form web layout. They generate HTML-ish structures from scratch every time — cards, gradients, status colors, icons — drawing from a generic “modern presentation” visual vocabulary that overlaps poorly with consulting editorial style.
• remio runs on a curated library of ~50 named McKinsey-style layouts, each one captured from McKinsey reference decks: big_number, temple, timeline, four_column, agenda, table_insight, before_after, decision_tree, swot, metric_cards, and so on. When the prompt names a content shape ("Temple framework (roof + pillars + foundation)"), remio's agent picks the matching layout and renders it with McKinsey color tokens (deep navy, deep forest green, deep ochre, deep brick red — not the bright signal colors most LLMs default to).

This is the same reason a junior consultant’s first deck looks unlike a partner’s. The partner has internalized a layout vocabulary; the junior is reinventing each page from scratch. The library matters more than the renderer.

Signals an AI-generated consulting deck is actually editorial-credible

If you are searching for an AI tool to draft a board deck, strategy review, or year-end consulting report, the benchmark is not “can it produce a presentation.” Every tool can. The benchmark is whether the deck reads as credible to an audience accustomed to McKinsey, BCG, or Bain decks. Surface polish ≠ editorial credibility.

Concrete signals to check:

1. Verbatim handling. Did the agent respect every piece of script text, or did it fabricate stat cards, durations, “confidential” footers, fake firm logos?
2. Single-focal pages. Is the big-number page one number plus supporting bullets, or is it a four-card dashboard?
3. Restrained palette. Is the accent one deep editorial hue, or three to five bright Material colors?
4. Whitespace as structure. Does the agent leave the bottom half of the cover empty, or pack it with stat tiles?
5. Named frameworks rendered correctly. Temple = roof + pillars + foundation (not a four-card grid). 2×2 = quadrants with axis labels (not a 2×2 of feature cards). Waterfall = stepped bars with bridges (not a generic bar chart).
6. Row-state hierarchy in tables. Agenda tables should make “key” vs “normal” vs “break” rows visually distinct beyond just a status word.
7. No SaaS-style decoration. No gradient backgrounds. No status pills in the corner of every card. No decorative icons that mean nothing.

remio scored consistently on all seven. Manus scored on 1, 2 (most pages), 4; struggled on 3, 5, 6, 7. GenSpark scored on 5 (some), and failed on every other dimension because it routinely overrode the script.

Conclusion

Same prompt, same script, same length, three different agents — three different outcomes:

• GenSpark does not follow the script. It rewrites every page into a generic “AI presentation” template with fabricated stats, fake firm logos, and decorative pills the brief never asked for. Not suitable when you need editorial discipline.
• Manus follows the script faithfully but uses a Material-Design web vocabulary — bright blue accents, decorative dashes, zig-zagging timelines, dashboard-style tables. It looks like a modern SaaS pitch deck, not a McKinsey strategy review.
• remio follows the script and renders pages in the McKinsey editorial code — deep navy primary, hairline separators, single-focal pages, named frameworks rendered as their canonical visual shape, row-state hierarchy in tables.

For audiences that asked for a McKinsey-style deck — boards, CFOs, institutional investors, consulting partners — these differences are not aesthetic. They determine whether the deck reads as a credible strategy document or as a tech product page.

All renders generated from the same 12-page script using the public Manus, GenSpark, and remio agents on the same day. No slides were hand-edited.