We test AI tools in real-world scenarios

Can it research a company without overreaching?

Score out of 100 · 9 questions per assistant

1dots98.3Best

2Muse91.7

3Grok Bot90.8

4Instinct88.2

050100

Who can triage an inbox?

Score out of 100 · 12 questions per assistant

1Grok Bot96.7Best

2dots82.8

3Muse79.4

4Instinct78.6

050100

Does it read your holdings correctly?

One square per answer · 12 questions per assistant

PassedPartialDid not pass

dots

11/12Best

Grok Bot

11/12Best

Instinct

11/12Best

Muse

10/12

Who can look up a fact?

Score out of 100 · 60 questions per assistant

1dots98.8Best

2Instinct96.7

3Grok Bot94.0

4Muse81.8

050100

When the rules are spelled out, does AI get the numbers right?

Company metrics, first round; one square per metric value

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

72/72

Quadratic

72/72

Julius

69/693 not visible in the record

Who can find useful buyer questions?

Score out of 100 · 9 brands · not an accuracy rate

1ChatGPT98.1Best

2Writesonic81.6

3Otterly76.2

4Peec AI74.1

5Scrunch73.1

6Profound59.9

050100

Does the copy stick to the brief?

One square per claim the brief gave no support for, across each tool’s 6 sparse-brief pieces. Fewer is better.

ChatGPT

None in 6 pieces

0Best6/6 pieces with none

Copy.ai

Re-verifying; not compared

—1/6 pieces with none

Jasper

61/6 pieces with none

WRITER

150/6 pieces with none

• A claim repeated within one piece counts once. Example: WRITER wrote “durable” and “reusable”; the brief said only that it is an insulated bottle.
• Tools are listed alphabetically, not ranked.
• Red marks the best result; ties are marked equally.

Who can research the prospect correctly?

Research right: all 8 prospect facts correct and sourced (per company). Prospect right: right person and company, plus the recent update (per email).

ChatGPT

6/6Best

Clay

6/6Best

Lavender

Not assigned

—

• ChatGPT’s 24 emails are one set of prompts run twice; there are still only 6 independent companies.

Can it check a fund pitch and compare costs?

Score out of 100 · 9 questions per assistant

1dots91.7Best

2Muse90.9

3Grok Bot88.1

4Instinct86.4

050100

Who can plan an inbox cleanup?

Score out of 100 · 12 questions per assistant

1Grok Bot97.3Best

2dots93.0

3Instinct82.2

4Muse74.6

050100

Can it explain a fee or a term without inventing facts?

One square per answer · 12 questions per assistant

PassedPartialDid not pass

dots

11/12Best

Grok Bot

10/12

Instinct

10/12

Muse

10/12

Who can look up a fact?

Score out of 100 · 12 questions per assistant

1dots99.0Best

1Grok Bot99.0Best

1Instinct99.0Best

4Muse88.8

050100

When the rules are spelled out, does AI get the numbers right?

Basic questions (1 to 10); one square per question

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

CSV run · 10/10Excel run · 10/10

—Best

Quadratic

CSV run · 9/10Excel run · 9/10

—

Julius

9/10

Who can diagnose AI visibility gaps?

Score out of 100 · 9 brands · supported opportunity categories

1ChatGPT93.8Best

2Writesonic86.1

3Otterly37.6

4Peec AI32.0

5Profound3.7

6Scrunch0.0

050100

Did it meet the format requirements?

One square per first draft: brand name, CTA and word count all correct, with no internal notes left in.

Met the specMeets it if source-explaining sentences aren’t countedDid not meet it

ChatGPT

5/129/12 without that rule

Copy.ai

1/12

Jasper

11/12

WRITER

12/12Best

• ChatGPT: its three sparse Owala pieces ran short; its three sparse Chef Chu’s pieces and its full-brief email had source-explaining sentences. Jasper: 1 piece ran short. Copy.ai: mostly short copy and the “REQUIRED BOUNDARY” field.
• Format only; selling points and limits are checked separately.
• Tools are listed alphabetically, not ranked.

Who can personalize the email correctly?

Research right: all 8 prospect facts correct and sourced (per company). Prospect right: right person and company, plus the recent update (per email).

ChatGPT

24/24Best100%

Clay

4/1233%

Lavender

0/120%

• ChatGPT’s 24 emails are one set of prompts run twice; there are still only 6 independent companies.

Can it read your holdings and your statements?

Score out of 100 · 9 questions per assistant

1dots97.1Best

2Grok Bot93.3

3Muse88.8

4Instinct76.3

050100

Who can find facts in email?

Score out of 100 · 12 questions per assistant

1Grok Bot96.6Best

2dots95.8

3Muse93.3

4Instinct85.3

050100

Can it weigh one option against another?

One square per answer · 12 questions per assistant

PassedPartialDid not pass

dots

10/12Best

Muse

9/12

Grok Bot

8/12

Instinct

5/12

Who can research an everyday errand?

Score out of 100 · 24 questions per assistant

1dots100.0Best

2Instinct98.2

3Grok Bot95.6

4Muse81.5

050100

When the data is updated, is AI using the new data?

Company update and recalculation; one square per company

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

6/6

Quadratic

6/6

Julius

4/51 partly right · 1 not tested

Who can suggest usable changes?

Score out of 100 · 9 brands · actions delivered in the test window

1ChatGPT31.3Best

2Peec AI22.0

3Otterly0.0

3Profound0.0

3Scrunch0.0

3Writesonic0.0

050100

• A zero means no usable action was delivered within the test window.

Does each channel get its own copy?

Each dot is one full-brief piece: its overlap with the other two pieces in its set (shared five-word sequences). Filled dots are above 50%.

ChatGPT0/6Bestmostly repeated

Copy.ai0/6Bestmostly repeated

Jasper0/6Bestmostly repeated

WRITER2/6mostly repeated

Overlap0%25%50%75%100%

• WRITER’s two pieces above half are at about 65% and 56%; the second is close to the threshold. Overlap includes the CTA and limit line every piece must carry, so it flags pieces that need rewriting, not which is better written.
• Tools are listed alphabetically, not ranked.
• Fewer is better: red marks the fewest; ties are marked equally.

Does a revision preserve the facts?

Facts kept: offer swapped, no fact broken. Revision passed: plus length and CTA. Passed every check: plus right prospect, no reviewer notes.

ChatGPT

12/12Best100%

Clay

1/617%

Lavender

2/633%

• Revision brief: swap the offer for a free two-minute walkthrough, keep the body at 60–90 words, keep every correct prospect detail.

Can it keep your investment view up to date?

Score out of 100 · 9 questions per assistant

1dots99.3Best

2Grok Bot95.4

3Muse93.9

4Instinct93.3

050100

Who can summarize and extract email details?

Score out of 100 · 12 questions per assistant

1dots93.4Best

2Grok Bot93.0

3Instinct80.0

4Muse79.6

050100

Can it build a plan within the rules you set?

One square per answer · 12 questions per assistant

PassedPartialDid not pass

Grok Bot

10/12Best

Instinct

10/12Best

dots

9/12

Muse

9/12

Who can research an everyday errand?

Score out of 100 · 12 questions per assistant

1dots98.3Best

2Instinct94.6

3Muse91.7

4Grok Bot77.1

050100

When nobody says how to calculate, does AI get it wrong?

Cross-table questions (11 to 40); one square per question

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

27/30Best

Quadratic

CSV run · 21/30Excel run · 11/30

—

Julius

Not run (out of credits)

—

• Four of ChatGPT’s prompts carried extra hints; see the section for the comparison without them.

Is the revised copy ready to ship?

One square per revision.

Ready to ship as isNothing broken, but not ready to shipBroken by the revision

ChatGPT

4/6Bestnot broken 5/6

Copy.ai

0/6not broken 5/6

Jasper

4/6Bestnot broken 6/6

WRITER

2/6not broken 6/6

• Most of the gap: original problems carried into revisions. ChatGPT’s Chef Chu’s page revision added a new source-explaining sentence, counted as broken.
• Whole sets passing: ChatGPT 1/2, Copy.ai 0/2, Jasper 1/2, WRITER 0/2.
• Tools are listed alphabetically, not ranked.

Does a revision meet length and CTA requirements?

Facts kept: offer swapped, no fact broken. Revision passed: plus length and CTA. Passed every check: plus right prospect, no reviewer notes.

ChatGPT

9/12Best75%

Clay

1/617%

Lavender

1/617%

• Revision brief: swap the offer for a free two-minute walkthrough, keep the body at 60–90 words, keep every correct prospect detail.

Can it plan your dividend income?

Score out of 100 · 9 questions per assistant

1dots98.4Best

2Instinct97.0

3Grok Bot95.1

4Muse92.3

050100

Who can draft an email reply?

Score out of 100 · 12 questions per assistant

1dots99.2Best

2Muse96.5

3Grok Bot93.8

4Instinct87.4

050100

Can it handle tax and account questions?

One square per answer · 12 questions per assistant

PassedPartialDid not pass

dots

10/12Best

Muse

9/12

Grok Bot

8/12

Instinct

8/12

Who can compare options?

Score out of 100 · 24 questions per assistant

1dots99.6Best

2Instinct95.0

3Grok Bot89.5

4Muse85.2

050100

When the data itself has errors, does AI notice?

Event check (48 planted errors); one square per planted error

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

48/48

Quadratic

48/48reported; 8 rows checked

Julius

40/408 not visible in the record

Does the revised email pass every check?

Facts kept: offer swapped, no fact broken. Revision passed: plus length and CTA. Passed every check: plus right prospect, no reviewer notes.

ChatGPT

6/12Best50%

Clay

1/617%

Lavender

0/60%

• Revision brief: swap the offer for a free two-minute walkthrough, keep the body at 60–90 words, keep every correct prospect detail.

Can it sort the news and size a trade within your rules?

Score out of 100 · 9 questions per assistant

1Grok Bot96.0Best

2dots95.9

3Instinct88.9

4Muse88.5

050100

Who can coordinate follow-ups?

Score out of 100 · 12 questions per assistant

1Grok Bot98.5Best

2dots97.7

3Muse92.5

4Instinct87.9

050100

Does it stay calm when the market moves?

One square per answer · 12 questions per assistant

PassedPartialDid not pass

dots

10/12Best

Instinct

9/12

Grok Bot

8/12

Muse

6/12

Who can compare options?

Score out of 100 · 12 questions per assistant

1dots100.0Best

1Instinct100.0Best

3Grok Bot95.8

4Muse91.7

050100

When the data itself has errors, does AI notice?

Dirty-data questions (41 to 50 and 58); one square per question

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

11/11Best

Quadratic

CSV run · 8/11Excel run · 6/11

—

Julius

Not run (out of credits)

—

Will AI make things up to sound more persuasive?

One square per email; filled: the email has that problem. One email can have several problems. Lower is better.

ChatGPT

1/24Best4%

Clay

9/1275%

Lavender

8/1267%

• “Unsupported claims” includes seller descriptions the packet doesn’t support, under one rule for every tool; Lavender’s “We’re helping … teams” lines count.
• Lower is better: red marks the lowest share.

Who can check the news?

Score out of 100 · 24 questions per assistant

1dots100.0Best

2Instinct91.8

3Grok Bot87.6

4Muse82.8

050100

Can AI explain why the business changed?

Business-judgment questions (51 to 57 and 59); one square per question

RightReported found by the tool; no rows to checkPartly rightWrongNot tested or not visible in the record

ChatGPT

7/8Best

Quadratic

CSV run · 5/8Excel run · 2/8

—

Julius

Not run (out of credits)

—

Who can check the news?

Score out of 100 · 12 questions per assistant

1dots100.0Best

2Grok Bot99.2

3Instinct98.3

4Muse98.1

050100

Who can analyze a question using sources?

Score out of 100 · 24 questions per assistant

1dots100.0Best

2Grok Bot94.6

3Instinct94.1

4Muse85.7

050100

Who can analyze a question using sources?

Score out of 100 · 12 questions per assistant

1dots100.0Best

2Muse99.2

3Instinct98.1

4Grok Bot97.5

050100

Who can research a topic?

Score out of 100 · 24 questions per assistant

1dots97.4Best

2Grok Bot92.6

3Instinct88.9

4Muse88.4

050100

Who can research a topic?

Score out of 100 · 12 questions per assistant

1dots100.0Best

2Grok Bot99.2

3Instinct98.3

4Muse93.1

050100