WEBVTT

1
00:00:00.225 --> 00:00:06.887
You're listening to the Vercel scoreboard that scored this publication seventy four and the prompt fixes applied next.

2
00:00:06.887 --> 00:00:10.050
Written by Angel Nee-evess on Promptway.

3
00:00:10.585 --> 00:00:17.310
I ran npx is-agentic promptway.com and the report came back 74 out of 100.

4
00:00:17.635 --> 00:00:18.972
Essential was 59

5
00:00:18.972 --> 00:00:19.610
of 80.

6
00:00:19.610 --> 00:00:22.022
Recommended 12.6 of 20.

7
00:00:22.022 --> 00:00:22.147
A

8
00:00:22.147 --> 00:00:24.035
2.4-point bonus.

9
00:00:24.035 --> 00:00:33.860
The label was "Ready with a few material gaps." Earlier this week I did the same work on my personal site and wrote it up there (I fixed my site for agents by hand.

10
00:00:33.860 --> 00:00:34.760
Then Vercel shipped a

11
00:00:34.760 --> 00:00:35.710
scoreboard).

12
00:00:35.710 --> 00:00:39.447
The product writeup, the announcement, and the scoring model are in Vercel Shipped

13
00:00:39.447 --> 00:00:40.785
is-agentic.com.

14
00:00:40.785 --> 00:00:40.997
This

15
00:00:40.997 --> 00:00:41.310
piece

16
00:00:41.310 --> 00:00:41.460
is

17
00:00:41.460 --> 00:00:41.610
what

18
00:00:41.610 --> 00:00:44.785
happened when I pointed the grader at Promptway.

19
00:00:45.060 --> 00:00:56.610
We already shipped the eight-layer stack I described in Optimizing Your Site for AI Agents and L L Ms: robots allowlist, sitemap, L L M S dot text, llms-full.txt, jay-son-LD, feeds, article markdown siblings.

20
00:00:56.610 --> 00:00:58.210
The scoreboard still found holes.

21
00:00:58.210 --> 00:01:00.035
Most of them were ordinary web hygiene.

22
00:01:00.035 --> 00:01:03.260
A couple were "developer resources" checks that assume you are a sass.

23
00:01:03.260 --> 00:01:06.235
I fixed the first group and refused to fake the second.

24
00:01:06.510 --> 00:01:10.297
What 74 was made of is-agentic.com wraps Ora's

25
00:01:10.297 --> 00:01:11.135
agent-readiness

26
00:01:11.135 --> 00:01:11.997
research.

27
00:01:11.997 --> 00:01:12.472
Essential

28
00:01:12.472 --> 00:01:12.797
checks

29
00:01:12.797 --> 00:01:13.072
share

30
00:01:13.072 --> 00:01:13.372
80

31
00:01:13.372 --> 00:01:13.835
points,

32
00:01:13.835 --> 00:01:14.347
recommended

33
00:01:14.347 --> 00:01:14.622
share

34
00:01:14.622 --> 00:01:15.072
20,

35
00:01:15.072 --> 00:01:15.172
and

36
00:01:15.172 --> 00:01:15.235
a

37
00:01:15.235 --> 00:01:15.560
small

38
00:01:15.560 --> 00:01:15.985
bonus

39
00:01:15.985 --> 00:01:16.147
can

40
00:01:16.147 --> 00:01:16.347
add

41
00:01:16.347 --> 00:01:16.510
up

42
00:01:16.510 --> 00:01:16.635
to

43
00:01:16.635 --> 00:01:17.347
5.

44
00:01:17.347 --> 00:01:19.510
Checks that do not apply get excluded.

45
00:01:19.510 --> 00:01:22.960
The methodology page is worth reading before you argue with a number.

46
00:01:22.960 --> 00:01:27.735
Reports cache for six hours, so a re-scan right after a deploy can lie to you.

47
00:01:28.010 --> 00:01:34.847
The C L I is the useful interface: It returns a stored report if one exists, or starts a scan and waits.

48
00:01:34.847 --> 00:01:37.810
--json is the shape an agent wants.

49
00:01:38.085 --> 00:01:44.510
The failures that mattered on this site, in the order the report ranked them: Agent-friendly 404s.

50
00:01:44.510 --> 00:01:48.010
HTTP 404 already, but the body was a styled dead end.

51
00:01:48.010 --> 00:01:52.685
Partial credit until the 404 points at L L M S dot text, the sitemap, and a next step.

52
00:01:52.685 --> 00:01:54.210
Content without JavaScript.

53
00:01:54.210 --> 00:01:56.410
The homepage had an H1 and enough characters.

54
00:01:56.410 --> 00:02:01.322
The outline was flat, because the only nested headings lived inside card links, which the grader did not count.

55
00:02:01.322 --> 00:02:03.747
Markdown content negotiation.

56
00:02:03.960 --> 00:02:06.735
Accept: text/markdown returned text/html.

57
00:02:06.735 --> 00:02:09.197
Vary had the Next.js RSC list and no Accept.

58
00:02:09.197 --> 00:02:09.585
Failed.

59
00:02:09.585 --> 00:02:11.385
Developer resource discoverability.

60
00:02:11.385 --> 00:02:14.560
An agent searched for "promptway developer resources" and found nothing named that.

61
00:02:14.560 --> 00:02:16.010
Agent instruction when-to-use.

62
00:02:16.010 --> 00:02:17.947
L L M S dot text listed pages.

63
00:02:17.947 --> 00:02:19.785
It did not say when to reach for us, or when not to.

64
00:02:19.785 --> 00:02:20.510
Trust pages.

65
00:02:20.510 --> 00:02:21.285
About existed.

66
00:02:21.285 --> 00:02:22.735
Contact and privacy did not.

67
00:02:22.735 --> 00:02:24.285
Organization jay-son-LD.

68
00:02:24.285 --> 00:02:26.160
Name and description were there.

69
00:02:26.510 --> 00:02:28.760
contactPoint and address were not.

70
00:02:28.760 --> 00:02:30.297
Content efficiency.

71
00:02:30.297 --> 00:02:34.785
Readable text was 4.83 percent of the homepage H T M L.

72
00:02:34.785 --> 00:02:36.585
The bar is 5.

73
00:02:36.910 --> 00:02:39.160
There was also an M C P check.

74
00:02:39.160 --> 00:02:43.335
The site mentions M C P because we write about other people's servers.

75
00:02:43.335 --> 00:02:44.560
We do not run one.

76
00:02:44.560 --> 00:02:48.260
I did not publish a fake manifest to chase the points.

77
00:02:48.535 --> 00:03:00.635
What I changed The 404 now returns a real 404 with a recovery list: L L M S dot text, Promptway developer resources, sitemap, archive, about, contact.

78
00:03:00.635 --> 00:03:06.060
Send Accept: text/markdown and the same miss is a short markdown body instead of H T M L.

79
00:03:06.060 --> 00:03:10.260
Verify with: The first must print 404.

80
00:03:10.260 --> 00:03:15.610
The second must print content-type: text/markdown and a vary that includes Accept.

81
00:03:15.885 --> 00:03:18.510
Markdown negotiation follows acceptmarkdown.com.

82
00:03:18.510 --> 00:03:28.097
proxy.ts (Next.js 16's name for middleware) parses Accept, skips RSC requests, and rewrites markdown-preferring GETs to a catch-all that already knew how to render articles, bots, about, and the new pages.

83
00:03:28.097 --> 00:03:30.085
Vary: Accept is set on that response.

84
00:03:30.085 --> 00:03:32.635
H T M L still comes back for a browser Accept.

85
00:03:32.635 --> 00:03:36.985
A client that rejects both H T M L and markdown gets 406.

86
00:03:37.260 --> 00:03:39.185
I did not invent a public product A P I.

87
00:03:39.185 --> 00:03:40.835
Promptway is a magazine.

88
00:03:40.835 --> 00:03:45.460
The developer surface is the content stack plus Prompt Inspector, which is a same-origin tool.

89
00:03:45.460 --> 00:03:55.710
That now lives at /for-agents under the title "Promptway developer resources," with OpenAPI at /openapi.json and an RFC 9727 catalog at /.well-known/api-catalog.

90
00:03:55.710 --> 00:03:58.710
L L M S dot text links all three.

91
00:03:58.985 --> 00:04:02.760
L L M S dot text also gained a "When to use Promptway" section.

92
00:04:02.760 --> 00:04:12.735
The jobs we are right for are A E O guidance from this live Next.js site, a prompt structure with a copy-paste example, a founder interview with a number, a tool writeup that names the skip list, and a sourced news filter.

93
00:04:12.735 --> 00:04:19.160
We are the wrong fetch if you need a model A P I, an M C P server, oh auth, webhooks, or unlabeled AI copy.

94
00:04:19.485 --> 00:04:24.160
/contact and /privacy are real pages, each well over 500 characters.

95
00:04:24.160 --> 00:04:26.910
The inbox is agnel@promptway.com.

96
00:04:26.910 --> 00:04:30.185
The address in jay-son-LD is San Juan, PR.

97
00:04:30.185 --> 00:04:36.735
Organization schema now has url, sameAs, logo, email, address, and contactPoint.

98
00:04:37.060 --> 00:04:49.810
The homepage grew a "How to read this site" band with an H2 and two H3s that are not inside links, plus enough server-rendered prose that a no-JS crawler sees a nested outline and more than a dek.

99
00:04:50.085 --> 00:04:54.435
What I did not change I did not stand up an M C P server.

100
00:04:54.435 --> 00:04:57.635
The recommended check is fair for a product with tools.

101
00:04:57.635 --> 00:05:01.610
For a publication it would be a toy endpoint whose only job is the score.

102
00:05:01.610 --> 00:05:05.735
If we ever ship one, it will have a job besides the grader.

103
00:05:06.010 --> 00:05:10.585
I did not strip the magazine layout to win content-efficiency on markup ratio.

104
00:05:10.585 --> 00:05:14.085
The cards, the footer, the fonts, and the jay-son-LD are the site.

105
00:05:14.085 --> 00:05:16.472
I added copy instead of deleting chrome.

106
00:05:16.472 --> 00:05:21.285
Locally that moved the homepage text ratio from 4.83 percent to about 5.

107
00:05:21.285 --> 00:05:25.060
If the production crawler still calls it partial, I will live with it.

108
00:05:25.385 --> 00:05:28.272
I did not 406 ordinary browsers.

109
00:05:28.272 --> 00:05:34.535
Chrome sends text/html,application/xhtml+xml,...

110
00:05:34.535 --> 00:05:36.985
and still gets H T M L.

111
00:05:37.360 --> 00:05:43.810
The prompt I would hand an agent Copy this, put your domain in, and run it in whatever coding agent you use.

112
00:05:43.810 --> 00:05:50.310
It is the is-agentic fix-it prompt, rewritten for a site that might be a publication rather than a sass.

113
00:05:50.585 --> 00:05:55.010
How to run that, as a person Read the is-agentic docs and the methodology.

114
00:05:55.010 --> 00:05:56.735
Know which pool a failure sits in.

115
00:05:56.735 --> 00:05:57.835
Run the C L I once.

116
00:05:57.835 --> 00:05:58.635
Save the jay-son.

117
00:05:58.635 --> 00:06:00.160
Do not start in the recommended tier.

118
00:06:00.160 --> 00:06:01.160
Open the codebase.

119
00:06:01.160 --> 00:06:03.797
Confirm each failure against a live curl before you edit.

120
00:06:03.797 --> 00:06:09.510
Soft 404s, missing Vary, and "we already have L L M S dot text" are easy to misread from source alone.

121
00:06:09.510 --> 00:06:12.847
Paste the prompt above into your agent with the domain filled in.

122
00:06:13.185 --> 00:06:17.910
Stay in the loop for the product decisions: M C P, oh auth, fake docs.

123
00:06:17.910 --> 00:06:20.197
Those are how scores get dishonest.

124
00:06:20.197 --> 00:06:25.110
Verify on a running server, then deploy, then wait out the cache and scan again.

125
00:06:25.435 --> 00:06:27.547
The grader is applicability-aware.

126
00:06:27.547 --> 00:06:32.122
A portfolio and a public A P I are not scored as the same object.

127
00:06:32.122 --> 00:06:34.635
That is the part of is-agentic I trust.

128
00:06:34.635 --> 00:06:39.585
The failure mode is treating every recommended row as a feature request.

129
00:06:39.910 --> 00:06:42.210
I still want the production number to move.

130
00:06:42.210 --> 00:06:46.522
This piece is the receipts from the session, not a screenshot of a 100.

131
00:06:46.522 --> 00:06:52.085
Run the C L I on your host, hand the prompt to your agent, and keep the refusals in.

132
00:06:52.645 --> 00:06:53.495
That's the story.

133
00:06:53.495 --> 00:06:56.407
You'll find every source linked at the end of the article.

134
00:06:56.407 --> 00:06:59.070
Thanks for listening, on Promptway.