How is the ranking calculated?

One score out of 100, weighted by capability 35%, reach 15%, ready to use 15%, autonomy 35%. Documented support earns one point, partial support half, and no support zero. Unknown evidence is excluded from the score denominator, with a separate research-completeness gate for awards.

Does AI visibility determine the winner?

No. The default order comes from the product score. Attention is measured separately using 100 neutral category-specific prompts across ChatGPT, Claude, Gemini and Google AI Overviews. Funding informs the attention verdict, never the product score.

What does overhyped mean here?

A product the market talks about far more than its score justifies, once the money behind it is accounted for. Attention and score are each converted to a position in the tracked field, a brand's disclosed funding is charged against its attention by up to 25 points, and anything still at least 15 points louder than it scores is called overhyped. A tool with almost no measured mentions never is — it is not loud, it is unmeasured.

What does underrated mean here?

The mirror of overhyped: a product that stands at least 15 points higher on the score than on attention once funding is charged against it, and that was named in at least one tracked answer. A tool that appeared in none is under the radar rather than underrated — there is no attention to judge it against yet. Underrated is the finding the page exists for — a product doing the work without the volume behind it — and it is the only verdict a tool can earn by being good rather than by being loud.

Free AI assistant ranking

The best AI assistants

58 AI assistants compared for hands-off work: what they can do, where you can use them, how much setup they need, and how much they do unattended.

Last measured 31 August 2026

This month's results

Our pick

Notis for hands-off work

The top 3

The highest-scoring award-eligible products. Rank badges match the full table.

Gold#1

Notis

Strongest on autonomy

90.6/ 100

Overall score

Silver#4

Zo Computer

Strongest on autonomy

86.3/ 100

Overall score

Bronze#5

Vellum

Strongest on ready to use

83.7/ 100

Overall score

Podium and bonus awards require at least 90% of the category matrix checked, at least 80% of every scored feature group checked, and a known setup level. Other scores are provisional. Unknown is not treated as no.

Bonus awards

Picks and standouts beyond the overall score. Attention combines visibility in AI answers with audience reach; gap awards also account for funding.

Strong score, less attention

Zo Computer

The biggest hidden gem: its score outpaces its attention, after accounting for funding.

Score rank: #4 of 58 · Attention rank: #20

More attention than score

Fyxer AI

The largest gap where attention outpaces the product score, after accounting for funding.

Score rank: #39 of 58 · Attention rank: #7

The ranking

Every tracked product by score, next to its documented capabilities, measured attention and disclosed funding. Sort or filter the table to explore the field. Provisional scores remain visible but cannot win awards.

Show

57 of 58 tracked assistants.

AI assistants ranked by the overall score, with the places each has moved since last week, documented capability, measured attention, disclosed funding, the movement on each since the last measurement and the verdict. Sortable by capability, attention or funding.
#	Assistant				Verdict
1	Notis Personal AI assistant	80—	0.8—	Bootstrapped—	Underrated
2	TownProvisional AI work assistant ("Townie") with built-in routine automation	80—	0—	$73M—	Under the radar
3	ChatGPTProvisional General-purpose AI assistant	90—	73—	$122B—	Earns it
4	Zo Computer Personal AI cloud computer	73—	0.4—	Undisclosed—	Underrated
5	Vellum Personal AI assistant	75—	2.7—	$25M—	Earns it
6	Tanka AIProvisional AI agent platform for startups with shared long-term memory (EverMemOS)	77—	0—	Undisclosed—	Under the radar
7	Lindy Slack-native AI teammate	61—	14.2—	$50M—	Earns it
8	WordwareProvisional AI assistant with a companion workflow-automation builder	57—	0—	$30M—	Under the radar
9	The LibrarianProvisional AI executive assistant for real estate agents	63—	0—	$2M—	Under the radar
10	Folk Crew AI personal assistant in your messaging apps	61—	0—	$6M—	Under the radar
11	Manus Autonomous general AI agent	62—	0.4—	$75M—	Earns it
12	OpenClaw Open-source self-hosted personal AI agent	88—	2.7—	Undisclosed—	Earns it
13	Atomic BotProvisional Managed hosting and installer for open-source personal AI agents (OpenClaw / Hermes)	52—	0—	Undisclosed—	Under the radar
14	Claude AI chat assistant	67—	30.3—	$125B—	Overhyped
15	Viktor AI employee for Slack and Microsoft Teams	53—	0.8—	$78M—	Earns it
16	MemoraeProvisional AI memory assistant in your messaging apps	52—	0—	$2M—	Under the radar
17	Quin AI admin assistant for meetings, CRM, and inbox follow-up	52—	0—	Undisclosed—	Under the radar
18	Prio AI chief of staff for founders and executives	47—	0—	Undisclosed—	Under the radar
19	Martin AI personal assistant reachable by phone, SMS, WhatsApp, email and Slack	41—	0—	$2M—	Under the radar
20	Junior AI employee and AI coworker platform	42—	0—	Undisclosed—	Under the radar

Score = capability 35%, reach 15%, ready to use 15%, autonomy 35%. How each component is measured.

Loud is not the same as capable

One chart, one line. Up is where a product stands on the score. Across is its measured AI-answer attention, adjusted for the money behind it, because money buys attention and a bootstrapped brand at the same volume has proved something a funded one has not. On the line, the attention is earned. The interesting ones are the distance off it.

Tracked productShut down or acquired

Both axes are a tool's position in the tracked field rather than its raw number, because a weighted score out of 100 and a long-tail attention measure cannot be subtracted from each other. Across is measured AI-answer attention plus the premium its funding explains, added to attention rather than taken off the score. Above the dashed diagonal a product is better than it is talked about; below it, the reverse. The shaded corridor is within 15 percentile points of the line, which is close enough to call even. The 23 assistants stacked at the left appeared in no tracked answer, raised nothing we could verify, so there is no signal left to tell them apart — they share a position because they share the measurement, and fanning them would invent an order. Every point is a row in the table above.

Does money buy attention?

The same field, split the other way: disclosed funding against measured attention. The dashed line is fitted to the products with known funding in this category; the distance from it shows which brands get more or less attention than that relationship suggests. Unknown funding is shown separately, not treated as zero.

Tracked productShut down or acquiredThe dashed line is the trend this month's field shows; the band is one standard deviation around it. The vertical axis is a square root, so the crowd under 5 stays readable next to the two brands above 30. The money axis stops at the decade that covers the field; a brand that has raised more is pinned to the right edge with a chevron and its real total, rather than stretching the axis until everyone else touches. Bootstrapped and undisclosed brands sit outside the model — a log axis has no zero, and an undisclosed round has no number — so they are fenced off at the left rather than dropped.

How the score is calculated

Every product is scored for hands-off work using the same documented feature evidence and weights.

35%Capability: Share of the capability grid the vendor documents. Partial support counts half.
15%Reach: Share of the places you might want it — WhatsApp, iMessage, Telegram, Slack, email, voice, web, phone.
15%Ready to use: Whether you can sign up and use it, or have to host it and bring your own model key.
35%Autonomy: Whether it works unattended: schedules, reminders, memory, a machine of its own and apps to act on.

Comparable research before awards

Provisional products remain in the full score-sorted table, clearly labelled. Rank badges on the podium retain their positions in that table.

Score ordering

Method hands-off-v2. Exact score ties break on autonomy, then capability, then slug. Popularity and funding never break a score tie.

Money raised, charged against attention

Attention is measured monthly, converted to a position in the tracked field, and then charged for the money behind it — up to 25 points, log-scaled and capped — before the field is ranked again on the result.

Money buys attention. A bootstrapped brand talked about as much as a funded one has proved something the funded one has not.

Overhyped and underrated are what is left: the distance between where a tool ranks on the score and where it ranks on attention after that charge. It is one number, and it is the one the quadrant above plots — the picture and the verdict are the same claim. The chart beside it shows the relationship the charge is drawn from.

The prompt panel

100 neutral prompts per category, fixed for the month, run through DataForSEO across ChatGPT, Claude, Gemini and Google AI Overviews — 400 answers per product. Each brand comes back as AI answer visibility and share of voice, which is the whole of its attention score.

A tool named in none of them scores zero on the panel — that is a finding about these 100 questions, not about the tool. A Google result with no AI Overview is a measured absence; a provider failure is unmeasured and blocks the whole month.

Each tool's own page lists the questions it appears in and its position on every one, so a score can be checked rather than trusted.

The capability matrix behind every score is open, cell by cell, in the AI assistants comparison directory.

Questions people ask about this ranking

How is the ranking calculated?: One score out of 100, weighted by capability 35%, reach 15%, ready to use 15%, autonomy 35%. Documented support earns one point, partial support half, and no support zero. Unknown evidence is excluded from the score denominator, with a separate research-completeness gate for awards.
Why is a score marked provisional?: Podium and bonus awards require at least 90% of the category matrix checked, at least 80% of every scored feature group checked, and a known setup level. Other scores are provisional. Unknown is not treated as no.
Does AI visibility determine the winner?: No. The default order comes from the product score. Attention is measured separately using 100 neutral category-specific prompts across ChatGPT, Claude, Gemini and Google AI Overviews. Funding informs the attention verdict, never the product score.
What does overhyped mean here?: A product the market talks about far more than its score justifies, once the money behind it is accounted for. Attention and score are each converted to a position in the tracked field, a brand's disclosed funding is charged against its attention by up to 25 points, and anything still at least 15 points louder than it scores is called overhyped. A tool with almost no measured mentions never is — it is not loud, it is unmeasured.
What does underrated mean here?: The mirror of overhyped: a product that stands at least 15 points higher on the score than on attention once funding is charged against it, and that was named in at least one tracked answer. A tool that appeared in none is under the radar rather than underrated — there is no attention to judge it against yet. Underrated is the finding the page exists for — a product doing the work without the volume behind it — and it is the only verdict a tool can earn by being good rather than by being loud.

Spotted something out of date?

Every cell here was read off a vendor’s own page on a dated pass, and vendors change their pages between passes. Tell us what moved, or which tool we are missing.

Stop reading rankings. Start delegating.

Message Notis from WhatsApp, iMessage, Telegram, Slack, or email and let it carry the work across every tool behind your business.

Start free