← AIAP Topic CoursesAll programsHomeSearch
AIAP Topic Courses·Using AI for Research·34:41

Evaluating AI Research Tools, Part 1: the Bake-off Method, a Market-Landscape Report, and a Notion Competitor Benchmark

Outskill instructor (unnamed) Research trainer - runs identical prompts through six AI research tools and scores them live
▶LMS lesson
← PreviousThe Universal Research Framework: SMART-AI Problems, the Research Tree, Three Source Layer
Lesson 3 of 4 in Using AI for Research
Next →Evaluating AI Research Tools, Part 2: Buyer Personas, Marketing Trends, a Content-Gap Anal

The short version

  1. No tool is best at everything - combine outputs and merge them with another LLM. Method: one prompt, unedited, across Claude Research, ChatGPT Deep Research, Gemini Deep Research, Manus, Genspark; score 1-5 on accuracy/sources (spot-check a cited claim), completeness, actionability. Subjective by design (0:00-0:03).
  2. Use case 1 - executive briefing on AI video tools (market size, CAGR, top 5-7 players, trends, regulation, investment, 1,500-2,000 words, cite everything): Claude pulled ~479 sources but synthesis was off; ChatGPT 8 minutes, ~22 sources; Gemini approves a plan first and gave the best tables; Manus exhaustive but messy; Genspark looked the part but sources could not be traced. A '67.4% share' claim was not in its cited page (0:03-0:28).
  3. Use case 2 - Notion vs Asana, monday.com, ClickUp, Airtable across features, pricing, segments, positioning, reviews, moves: Gemini best citations, weak formatting; Manus one markdown doc per question; Genspark most actionable presentation (0:26-0:34).

At a glance, three clicks deep

Skim here first: the closed row is the glance, open is the study card with the key points and timestamps, and the ↓ link drops to that concept's full write-up below.

01The bake-off method and rubricIdentical prompt;›

Identical prompt; three scores; combine.

Method (0:00-0:03)

↓ Full write-up of this concept

02Use cases 1 and 2: market landscape and competitor benchmarkGemini for depth and tables;›

Gemini for depth and tables; Genspark for looks; verify every stat.

Use case 1 (0:03-0:28)

Use case 2 (0:26-0:34)

↓ Full write-up of this concept

The concepts in full

01

The bake-off method and rubric

Same prompt, five tools, three scores. Then feed all five answers to a sixth model and let it reconcile them.

Run the identical prompt unedited in each tool; score 1-5 on accuracy and sources (open a cited page and check the claim is there), completeness and relevance, actionability and clarity; declare it subjective. The strategic advice: no single tool wins, so combine outputs and synthesise with another LLM.

Why it matters

A repeatable way to choose tools as they change.

02

Use cases 1 and 2: market landscape and competitor benchmark

Four hundred and seventy-nine sources and a muddled synthesis, versus twenty-two sources and a clean briefing. More is not better.

Prompt 1 (excerpt): 'comprehensive market landscape report on AI tools for video creation - market size and projected growth with revenue figures and CAGR, top 5-7 companies and positioning, trends, regulatory environment, investment activity; cite all sources; executive briefing for C-level; 1,500-2,000 words.' Claude: ~479 sources, synthesis 'a little off'; ChatGPT: 8 minutes, ~22 sources, solid; Gemini: shows a plan to approve, best tables and competitor coverage; Manus: exhaustive, poorly organised, rendering glitches, unlinked claims; Genspark: executive-briefing look, untraceable sources. Spot-check: a '67.4% share' sentence was absent from its cited page. Scores (accuracy/completeness/actionability): Claude ~3.5-4/3.5/3.5; ChatGPT 3.5/4/3.5; Gemini 4/4.5/4; Manus 3.5/4/3.5; Genspark 3/4/4.5. Prompt 2: Notion vs Asana, monday.com, ClickUp, Airtable - features, pricing, segments, positioning, channels, strengths/weaknesses, reviews, strategic moves. Gemini 4/4/3.5 (good citations, weak formatting); Manus 3.5/4/4 (separate docs); Genspark 3.5/4/4.5; Claude ~3.5-4/4/4; ChatGPT 4/3.5/3.5-4.

Why it matters

Two realistic business prompts with the scorecards attached.

This idea elsewherecontinuesVerdict in Part 2

Tools referenced

ToolCoverageMomentContext
ClaudedemonstratedClaude Research
ChatGPTdemonstratedDeep Research
GeminidemonstratedDeep Research
Manusdemonstrated
Gensparkdemonstrated
NotionmentionedBenchmark subject

Action items

    Resources mentioned

    Resources
    • docLesson page on the platform

    Extraction notes

    This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

    Transcript corrections applied

    The transcript saysThe trainer actually means
    Manas / MarisManus
    Genespark / GemsparkGenspark
    Cling / Soera / Hagen / v o 3Kling / Sora / HeyGen / Veo 3
    3.5243.5 to 4
    cloud / Cloud co work / cloud researchClaude / Claude Cowork / Claude Research
    chat g b t / ChargeGPT / Chargebee / chat GBTChatGPT
    ATN / n a 10 / Anytimen8n

    True on recording day — verify before relying

    ← PreviousThe Universal Research Framework: SMART-AI Problems, the Research Tree, Three Source LayerNext →Evaluating AI Research Tools, Part 2: Buyer Personas, Marketing Trends, a Content-Gap Anal