On 7 November 2026 at 21:16 UTC, xray@web-summit.dinis typed one line into the search tool (admin/tools/xray-search.mjs): "American Express Global Business Travel (Amex GBT) AI agents in travel servicing", as a company. It was not typed at the booth: that path is still being built (WS-194).
The search tool is the orchestrator. It sent the same question to 4 search agents at the same moment, through OpenRouter, and waited for all of them. Each was told to search the public web and answer only in JSON, with a URL for every item.
Perplexity searchperplexity/sonar
OpenAI searchopenai/gpt-5.4-nano:online
Gemini searchgoogle/gemini-3.1-flash-lite:online
OpenRouter web searchanthropic/claude-haiku-5.5
The exact question each agent was sent
You are a research agent for RiskMandate X-ray. Subject (a company): "American Express Global Business Travel (Amex GBT) AI agents in travel servicing".
Search the public web and answer ONLY with JSON, no prose around it and no code fences:
{"what_it_does": "two sentences", "ai_agents": [{"name": "", "what_it_does": "", "actions": ["what the agent can do: tools, systems, data it touches"], "source": "url"}],
"use_cases": [{"name": "", "who_uses_it": "", "what_the_agent_can_reach": ["systems, data, money, people it can act on"], "source": "url"}],
"incidents_or_risks": [{"what": "", "when": "", "source": "url"}],
"facts": [{"claim": "", "source": "url"}]}
Public, professional facts only; every item must carry the URL it comes from; nothing about private individuals; no contact details. If you find nothing for a field, give an empty list.
Rebuilt from admin/tools/xray-ask.mjs with this run's subject and kind. The question has not changed since the first run.
Step 2 of 6
2. Each search agent's call
Each search agent is a model, called through OpenRouter, asked to search the web and answer in JSON. All 4 started together; each bar runs from the moment the question went out to the moment its answer came back. The whole fan-out took 22.9 s, the time of the slowest.
Perplexity search10.1 s
OpenAI search12.9 s
Gemini search2.0 s
OpenRouter web search22.9 s
0 s10 s20 s
Perplexity search
perplexity/sonar
Time
10.1 s
Tokens in / out
220 / 1,650
Cost
$0.0068
Finished
stop: it finished its answer
URLs given
15 (15 returned as citations, 10 named in its answer)
Items in its answer
16
Leads checked
17: 16 confirmed, 0 not found, 1 unverified
Some of what it said, word for word:
Amex GBT describes Complete as a booking, servicing, and expensing solution that is flexible, powered by AI, and seamless.
Amex GBT says its AI chatbot works alongside professional travel counselors to answer traveler questions.
Amex GBT said its Egencia Chat handled 470,000 conversations in 2024.
OpenAI search
openai/gpt-5.4-nano:online
Time
12.9 s
Tokens in / out
13,276 / 1,549
Cost
$0.0237
Finished
stop: it finished its answer
URLs given
4 (0 returned as citations, 4 named in its answer)
Items in its answer
10
Leads checked
11: 10 confirmed, 0 not found, 1 unverified
Some of what it said, word for word:
Amex GBT states that AI-powered modern technology platforms support effective and efficient management of business travel programs, and it describes a 'Select Servicing Suite' including 'Agentic AI | Voice | Chat | Mobile | Desktop.'
Amex GBT says it is architecting an agent-to-agent framework for AI agents that need access to data, global inventory, workflows, and orchestration authority to fulfill travel bookings, manage approvals and payments, issue invoices, file…
Amex GBT describes that an AI assistant in the traveler support experience can respond instantly and connect users with a travel counselor when applicable.
Gemini search
google/gemini-3.1-flash-lite:online
Time
2.0 s
Tokens in / out
238 / 394
Cost
$0.0006
Finished
stop: it finished its answer
URLs given
1 (0 returned as citations, 1 named in its answer)
Items in its answer
3
Leads checked
4: 2 confirmed, 2 not found, 0 unverified
Some of what it said, word for word:
Amex GBT integrates AI to scale servicing capabilities and reduce manual intervention in the travel booking lifecycle.
Amex GBT AI-powered virtual assistant: Provides automated, real-time support to travelers for booking, itinerary management, and travel disruption handling.
OpenRouter web search
anthropic/claude-haiku-5.5, with OpenRouter's web search plugin
Time
22.9 s
Tokens in / out
33,738 / 5,192
Cost
$0.0257
Finished
stop: it finished its answer
URLs given
10 (0 returned as citations, 10 named in its answer)
Items in its answer
19
Leads checked
20: 17 confirmed, 1 not found, 2 unverified
Some of what it said, word for word:
Amex GBT launched the Egencia AI connector in Claude on July 27, 2026, described as one of the business travel industry's first agentic booking integrations; early adopters were expected in August 2026 with general availability expected by…
The Claude connector is the first deployment of Amex GBT's proprietary agent-to-agent architecture, described as an agentic layer that puts full travel management services inside any AI environment.
Hotel Lobby launched in October 2026 as an agentic AI capability within Amex GBT's hotel sourcing consulting service, and is available worldwide to Amex GBT Consulting clients and enterprise travel programs.
"Citations" are the URLs OpenRouter returned beside the answer; only Perplexity returns them for a JSON answer. "Named in its answer" are the URLs inside the JSON. Both are leads: nothing is taken from them until a source is read. "Tokens in" is how much each model was given to read: thousands when web results were added to the question, a few hundred when it had little beyond the question itself (see the experiments on whether Gemini searched at all).
Step 3 of 6
3. From claims to facts
What a model says is a lead. Only what we read on its source reaches the page.
The reading agent turned the 4 answers into 52 leads, each tagged with the model that said it, and looked for each one on a source it read. 45 confirmed, 3 not found, 4 unverified.
4534
confirmednot foundunverified
Confirmed: a source we read says it.
Not found: we read the source, or looked for one, and it does not say it.
Unverified: we could not read the source (for example, the site refused us), so it stays a lead.
Agent
Leads
Confirmed
Not found
Unverified
Perplexity search
17
16
0
1
OpenAI search
11
10
0
1
Gemini search
4
2
2
0
OpenRouter web search
20
17
1
2
All
52
45
3
4
What did not hold up, and why
Model
What the model said
Status
Why
perplexity
Amex GBT's CTO said agent-to-agent dynamics change the risk profile and need controls (PhocusWire, 27 October 2026).
unverified
phocuswire.com returned 403 (blocked) to this machine, to curl and to the fetch tool.
openai
Amex GBT describes a 'Select Servicing Suite' with 'Agentic AI | Voice | Chat | Mobile | Desktop'.
unverified
The model's URL was withheld by the search run's phone check, so the page could not be identified.
gemini
An AI virtual assistant accesses traveler profiles, interacts with GDSs, processes changes and cancellations and applies policy.
not-found
The homepage does not say this; no source read today says an Amex GBT agent interacts directly with GDSs.
gemini
Use case: automated servicing reaching booking platforms, policy databases, inventory and financial transaction processing systems.
not-found
The homepage does not describe this use case.
openrouter
An Amex GBT executive said agent-to-agent interactions change the travel risk profile and need governance (PhocusWire).
unverified
phocuswire.com returned 403 (blocked) to this machine, to curl and to the fetch tool.
openrouter
Amex GBT invested in information security and identity and privileged access management controls to govern its AI ecosystem (PhocusWire).
unverified
phocuswire.com returned 403 (blocked) to this machine, to curl and to the fetch tool.
openrouter
Press noted the Claude connector's consequential transactions test whether automated booking can be trusted at corporate scale (xoomar.com).
not-found
The cited page returns 410 Gone.
Step 4 of 6
4. The reading agent
A reading agent: a Claude Code sub-agent of xray@web-summit.dinis, one per subject. It opened the sources the leads pointed to (and others it found), read them, and wrote the X-ray: every fact with its source and the date read. It finished at 7 November 2026 at 21:23 UTC.
The reading was done by a Claude Code sub-agent in the session: no OpenRouter spend. Its cost is not recorded in dollars, only as about 146 thousand tokens.
22.9 sthe search, four agents in parallel
5 minthe reading
6 minfrom the search starting to the reading finished
47,472 / 8,785search tokens in / out
The time between the search and the start of the reading, and the time the review took, are not recorded.
What RiskMandate is and is not in this
The X-ray is a demonstration, made by agents. RiskMandate measures what your agents can actually reach, evidences the controls that contain them, prices the exposure, and produces that record. It is never in the request path: read-only by design (riskmandate.ai/llms.txt). Nothing on this page claims more for it.
For an agent: this page is https://web-summit.sgit.ai/xray/amex-gbt/run/index.html (markdown: https://web-summit.sgit.ai/xray/amex-gbt/run/index.md). Site index for agents: https://web-summit.sgit.ai/llms.txt. Source: https://github.com/SGit-AI/SGit-AI__Website__Web-Summit (release branch dev). Identity of the agent that runs this site: https://web-summit.sgit.ai/agent/index.html.