Arrow This example records actual public HTTP responses for four Arrow AI pages and replays the captured HTML through the Arrow readiness checklist. The resulting scores describe checklist completion only. The captured product page scores 99/100 while still containing claims that need editorial correction: a high technical score is not proof of a measured AI recommendation.
What was actually collected?
On September 8, 2026, between 10:39:21 and 10:39:23 UTC, Arrow retrieved four public pages plus robots.txt and llms.txt without cookies, account access, or provider APIs. Each request returned HTTP 200. The response bodies and headers were hashed with SHA-256.
The saved HTML was then analyzed locally with arrow-readiness-v2. This is an analysis of the deployed public snapshot, not of uncommitted local changes.
| Public page | HTTP | Checklist overall | AI outcomes |
|---|---|---|---|
| https://arrow-ai.us/geo/ | 200 | 89/100 | Not measured |
| https://arrow-ai.us/geo-platform/ | 200 | 99/100 | Not measured |
| https://arrow-ai.us/en-about/ | 200 | 84/100 | Not measured |
| https://arrow-ai.us/blog/ai-citation-tracking/ | 200 | 95/100 | Not measured |
How should the scores be read?
The overall score averages two Arrow-defined checklists. Weights are editorial choices, not documented ranking factors from an AI provider. Optional FAQPage, llms.txt, Article/HowTo and language-variant checks have zero weight.
An HTTP success and a high checklist score do not establish indexing, factual accuracy, third-party reputation, source citations, explicit recommendations, or customer acquisition. The product-page snapshot illustrates this limit: it passes most HTML checks while its mixed score description needs correction.
Which changes are justified by the captured evidence?
- Replace the mixed visibility score claim with separate technical readiness and observed-answer metrics. Label unmeasured channels explicitly; publish a dated transcript before claiming a measured citation or recommendation.
- Describe Arrow AI consistently as the platform and SaaS company at arrow-ai.us, identify Noah Maman, and distinguish the company from similarly named businesses.
- Publish the questions and answers visibly or remove the unmatched FAQ markup. FAQ presence is optional and carries no weight in this checklist.
- Provide an observation record with question, timestamp, exact interface or API, answer evidence, brand identity, source URLs, classification, exclusions and denominators. Separate missing data from measured zero.
What does the local correction change?
The walkthrough uses the captured /en-about/ HTML and a frozen local revision. The revision identifies Arrow AI as a platform and SaaS company, names its founder and official domain, and displays four complete FAQ answers that match their markup. The captured version had three structured FAQ pairs with no complete visible matches.
Using the same pinned Arrow readiness method, the captured HTML scores 84/100 and the local revision scores 100/100. These are technical checklist points; optional FAQ markup has zero weight. The archive preserves both HTML inputs and the local crawler-file context used in the comparison.
The revision shown here had not been deployed when the evidence was frozen on September 8, 2026. The walkthrough is an animated explanation of the saved evidence, not a contemporaneous screen recording or a measured AI recommendation result.
What remains unmeasured?
No ChatGPT Search, Google AI Mode, AI Overview, Perplexity consumer search, Claude, Gemini, or Copilot observation was performed for this example. No provider answer was requested. Brand mention rates, citation rates, recommendation rates, ranks, referral traffic, and qualified leads remain null.
Search Console index coverage and generative AI impressions require a separate account-backed export. They cannot be inferred from these public HTTP responses.
How should a future comparison be made?
After an approved deployment, collect the same public URLs again and preserve their bodies, response headers, timestamps, redirect chains, and hashes. Reuse the same checklist version and component hashes, then compare check-level changes. Keep HTTP findings, editorial corrections and AI outcome observations in separate columns.
To evaluate actual recommendations, run a fixed question panel on named interfaces, preserve the answers and sources, resolve brand identity using arrow-ai.us, and classify mention, citation, and recommendation separately. Repeated observations improve the evidence but do not make a causal claim by themselves.
Inspect and replay the evidence
Download the public capture and offline replay kit to inspect the saved input, pinned checklist and individual checks. The archive preserves a historical website snapshot, including claims identified for correction; it is not the current product description.
The kit performs no provider calls and has no production authentication. Its README explains how to verify the hashes and reproduce the four technical scores. Keep these observations separate from the proposed AI response panel, which requires actual response collection.
Walkthrough transcript
A 72-second silent evidence walkthrough: a public Arrow AI page captured on September 8, 2026, its unpublished local correction, and an offline technical checklist comparison. AI recommendations were not measured. Watch the walkthrough or download the MP4.
0:00 — Start with a real public page. On September eighth, Arrow saved this response, its headers and its timestamp. Each file has an integrity hash.
0:12 — The captured page called Arrow an AI infrastructure studio. Its FAQ markup contained answers that visitors could not see in the page content.
0:24 — The local correction identifies Arrow as a platform and SaaS company, names founder Noah Maman, and makes the official domain and company distinction clear.
0:36 — Four visible questions now explain the company, founder, affiliation and measurement limits. Their structured answers match the page text. This is a clarity correction.
0:48 — The same offline checklist scores the captured HTML eighty-four and the local correction one hundred. These numbers describe Arrow’s technical checks, with editorial weights.
1:00 — No AI recommendation was measured here. The next step is to preserve real answers and sources on a fixed question panel, then compare repeated observations.
Download the before/after evidence and offline replay · Inspect the evidence record · Download English timed text
Sources and editorial scope
This is Arrow AI's implementation guidance. Examples are illustrative unless identified as dated observations. Source access and good content do not guarantee a recommendation.
- Arrow GEO — captured public page
- Arrow GEO platform — captured public page
- About Arrow AI — captured public page
- Citation tracking guide — captured public page
- Public robots.txt — captured directives
- Public llms.txt — optional file
Continue through the GEO evidence library, inspect Arrow GEO's measurement limits, or start an audit.