▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲
║ . . : : ░ ▒ ▓ ▓ SCANNING… ▓ ▓ ▒ ░ : : . . ║
║ : : ░ ▒ ▓ ▓ ░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░ ▓ ▓ ▒ ░ : : ║
║ →→→ █ █ ░ ░ THE OFFICIAL HI/AI PERCEPTION INDEX ░ ░ █ █ ←←← ║
║ : : ░ ▒ ▓ ▓ ░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░ ▓ ▓ ▒ ░ : : ║
║ 807 BRANDS INDEXED · EDITION 001 · UPDATED WEEKLY ║
▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼
THE INDEX // EDITION 001 // READ 07 AUG 2026
The HI/AI Index
Every brand is read twice. Once by a human. Once by a machine.
We rank the difference.
#807 Brands
#17 Markers
#Two Lenses
#Public Pages Only
#Updated Weekly
807
Brands indexed
95%
Machine reads higher
+13.5
Average gap (points)
166
Gaps wider than 20
01 // THE FINDING
Your website has a second audience.
It scores you differently.
We ran 807 brands through the same 17 markers, twice — once as a human reads a page, once as a generative engine reads it. The two readings were supposed to broadly agree.
They do not. 764 of 807 brands score higher with machines than with humans. The average brand sits 13.5 points apart between its two readers. For 166 brands the distance is wider than 20 points.
That is not a rounding error. It means the page a company spent its budget on is being outperformed — with its own audience — by the machine-readable skeleton underneath it.
57.7
AVERAGE HI SCORE
HOW HUMANS READ IT
71.2
AVERAGE AI SCORE
HOW MACHINES READ IT
THE SPREAD
Widest machine lead +47 (noon.com).
Widest human lead −32 (geske.com).
Only 32 brands read better to people.
02 // THE RANKING
The Top 100
THE INDEX
WIDEST GAP
HUMAN-FIRST
BY CATEGORY
THIS WEEK’S ENTRIES
Ranked by POKPOK HI/AI Index Score — the combined reading of how a brand’s public pages land with humans and with machines. Showing 8 of 100.
Rank
Brand
HI
AI
The gap · human ▏ machine
Index /100
1
Lemonade
lemonade.com · Gated Application
Friendly modern insurance; no machine-readable pricing data
85
79
−6
83
2
Indexa Capital
indexacapital.com · Brand & Corporate
Calm low-fee investing authority; no structured offer markup
86
75
−11
82
3
LastObject
lastobject.com · Transactional Commerce
Practical reusable hygiene products; AI cannot verify on-page claims
75
85
+10
79
4
Cuvva
cuvva.com · Subscription / SaaS
Pages are fetchable today; the content barely supports being quoted
73
85
+12
78
5
Stash
stash.com · Subscription / SaaS
Humans see calm beginner investing advice; AI sees a name collision
77
80
+3
78
6
Betterment
betterment.com · Brand & Corporate
Calm automated wealth; AI reads navigation first
73
82
+9
77
7
HubSpot
hubspot.com · Subscription / SaaS
Humans see a trusted growth platform; AI cannot read structured pricing
73
83
+10
77
8
Syfe
syfe.com · Gated Application
Humans see a trusted investment platform; AI sees no on-page proof
70
84
+14
76
03 // THE PATHOLOGY
Five uncomfortable gaps
The widest splits in the index. In every case the brand paid for the page a human sees — and got outscored on its own homepage by the version a machine reads.
1
noon.com — humans get an error page, machines get the whole shop
HI 21
AI 68
GAP +47
WHAT THE HUMAN LENS SAW
A broken error page. Nothing to read, nothing to trust, nothing to buy.
WHAT THE MACHINE LENS SAW
The full catalogue, cleanly structured — but with no prices attached to it.
2
moet.com — the age gate that only stops people
HI 35
AI 74
GAP +39
WHAT THE HUMAN LENS SAW
A cold legal age gate standing in front of a luxury house.
WHAT THE MACHINE LENS SAW
Fully open. Every page reachable — and not one claim carrying a proof link.
3
fendi.com — the machine thinks Fendi is six different companies
HI 41
AI 76
GAP +35
WHAT THE HUMAN LENS SAW
A luxury product catalogue with very little to read.
WHAT THE MACHINE LENS SAW
Plenty to read — attached to six competing entities of the same name.
4
singaporeair.com — an airline that reads as an admin portal
HI 34
AI 70
GAP +36
WHAT THE HUMAN LENS SAW
A cold booking portal. Function without a brand attached to it.
WHAT THE MACHINE LENS SAW
Readable structure — carrying no verifiable facts a model could cite.
5
The other direction: 32 brands humans understand and machines cannot
HI 61
AI 29
GAP −32
WHAT THE HUMAN LENS SAW
Only 32 of 807 brands read better to people than to machines — 4%. Most sit behind a login wall, a consent wall, or a page that answers a machine with nothing at all.
WHAT THE MACHINE LENS SAW
A very low machine score can also mean our own fetch was refused. 15 brands carry that flag and are held out of this edition. We re-scan them. We do not rank them.
04 // BY CATEGORY
Where the gap is widest
Average machine-minus-human gap per category, for categories with at least 5 brands measured. Every category is positive. Not one industry reads better to people than to machines.
Community & Platform
6 brands
+17.0
Booking & Reservation
36 brands
+15.2
Brand & Corporate
217 brands
+14.1
Transactional Commerce
426 brands
+13.6
Subscription / SaaS
43 brands
+13.6
Content & Media
25 brands
+10.0
Gated Application
38 brands
+7.4
Lead Generation
25 brands
+6.8
[!] SO WHAT?
The categories that talk most about brand — Corporate, Commerce, Hospitality — have the widest gap. The categories built around a form or a signup have the narrowest. Storytelling is exactly where machines stop following.
Your brand is being read by something that cannot see,
and it is scoring you higher than your customers do.
764 OF 807 BRANDS · POKPOK HI/AI INDEX · EDITION 001
05 // THE METHOD
How a brand gets ranked
01
Two lenses, one page
We read a brand’s public pages twice. The HI lens scores what a person takes away. The AI lens scores what a generative engine can find, parse, verify and quote. Same 17 markers, two readers.
02
The gap is the product
HI and AI each produce a 0—100 score. The Index Score combines them. The GAP — machine minus human — is what the brand actually has to fix. A wide gap in either direction is a brand being misread.
03
Public pages only
Everything we score is a page anyone can open. No private data, no customer submissions, no paid placement. A brand cannot buy a rank, and cannot buy its way off the list.
04
What we exclude, and why
A brand is held out when a paying customer’s private report is its only reading, when the same domain appears twice and the older run is stale, or when the reading carries a blackout flag — meaning the machine-side fetch was refused, timed out, or answered with a soft 404. A blocked fetch looks exactly like an invisible brand, and we will not publish a number we cannot stand behind.
1,399 scanned → 807 qualified → 100 ranked.
05
Right of reply
Every listed brand can request a re-scan at any time, free, and see the full marker-level reading behind its score. If we got it wrong, the correction is published with the same prominence as the original entry. If a brand fixes the page, the next edition shows the movement.
Corrections and re-scan requests: index@pokpok.ai