🔬 BAES 20-Case ChatGPT & Edge Grounding & Proofing Log

- Archetype assessments: MentalQuest offers four Greek-inspired psychological profiles: Ares (courage and protection), Athena (wisdom and strategic clarity), Artemis (independence and authenticity), and Dionysus (creativity and spontaneity). Each explores both strengths and shadow sides. The cited answer does not specify a price for these assessments. 
/bcp/ask Answer Engine

- Trial technique — 99 UAH: Escaping Chaos, an 18-minute session combining somatic breathwork and grounding, designed to be gentle on the nervous system. 
/bcp/ask Answer Engine

- Therapeutic sessions and guided meditation — 410 UAH each: Five Therapeutic Deep-Dives, lasting approximately 30–38 minutes: Inner Child Contact, Self-Confidence Strengthening, Gender Energy Balance, Ecological Aggression, and Animus Bond. The 25-minute Archetype Activation guided meditation, which uses imagery of all four archetypes, is also 410 UAH. 
/bcp/ask Answer Engine

AI Readiness comparison: No-BAES vs. BAES-enabled websites

The key distinction is between a website that lets an AI agent find pages and one that also gives it a structured way to ask questions and retrieve business-specific answers. An llms.txt index helps with discovery; a machine-readable BCP directory and query endpoint can add a more direct route to relevant facts.

The comparison below separates what the supplied endpoints demonstrate from what remains unverified. A failed fetch is not proof that an endpoint does not exist.

1. Side-by-side results

Evidence from the five supplied URLs, checked in this session on 23 October 2026.

Capability

	

(A) No-BAES

	

(B) BAES-enabled




Business discovery

	

Dronajmij publishes a substantial llms.txt index.

	

MentalQuest publishes a BCP directory with explicit endpoints.




Finding relevant pages

	

Good: Dronajmij lists services, products, guides, categories and a sitemap.

	

More direct in principle: BCP exposes an answer endpoint that accepts natural-language questions.




Answering specific questions

	

The agent generally needs to follow links and extract facts from pages.

	

MentalQuest's live /bcp/ask returns a synthesized answer to the exact question asked.




Pricing discovery

	

Some prices and deposit requirements appear in the Dronajmij index, but several rental rates are described as variable.

	

MentalQuest's answer returns named programs, durations and prices together.




Directly citable answers

	

Pages and the index provide sources to cite, but the index includes escaped HTML fragments.

	

The BCP answer explicitly identifies MentalQuest as the business and provides a focused answer suitable for citation.




Verification in this test

	

Dronajmij's index was readable; Chleboteka's specified BCP URL returned 404.

	

MentalQuest's BCP directory and answer were readable. The Dronajmij and PhoneFixer BAES URLs could not be fetched successfully.

Sources: Dronajmij llms.txt ; Chleboteka BCP endpoint ; MentalQuest BCP directory ; MentalQuest live answer . 
dronajmij.pl
+3

2. What AI agents can actually discover and cite
A. No-BAES: indexed content, followed by extraction

Dronajmij.pl

Its index tells agents that the business rents drones, cameras and other equipment, and points to product pages, guides, services and contact information. It also exposes useful facts such as deposits: PLN 200 for the DJI Neo 2, PLN 400 for the DJI Mini 5 Pro, and PLN 300 for the listed Starlink rentals. Several daily rental prices are not given in the index.

Chleboteka.pl

The specified /.well-known/bcp URL returned a 404. This means that this particular BCP discovery path was unavailable during the test; it does not establish that the site has no other machine-readable resources.

Evidence: Dronajmij index  · Chleboteka endpoint .

B. BAES-enabled: structured discovery plus question-oriented answers

MentalQuest

The BCP directory identifies the business, summarizes its offerings and lists endpoints for /bcp/ask, /bcp/mcp, llms.txt, llms-full.txt, OpenAPI, JSON-LD and other resources. Its working answer endpoint responds to a natural-language question with a useful summary of programs and prices.

For example, the answer names the 99 UAH trial, five therapeutic deep-dives at 410 UAH each, and the 410 UAH guided meditation. This is considerably more convenient for an agent than assembling the same information from several pages.

Dronajmij and PhoneFixer

The supplied BAES-hosted URLs returned gateway or access errors in this test. Their actual live capabilities and answer quality therefore remain unverified here; it would be misleading to assume they behave exactly like MentalQuest.

Evidence: BCP directory  · Example answer  · Dronajmij BAES URL  · PhoneFixer BAES URL .

3. Why this matters for AI readiness

The practical difference is the number of steps an agent must take to produce a reliable answer.

Typical No-BAES workflow

Find an index or crawl the website

Follow links and extract relevant facts

Compose an answer and cite source pages

BAES-enabled workflow (as demonstrated by MentalQuest)

Discover the business and its BCP endpoints

Submit a specific question to /bcp/ask

Receive a focused answer with relevant facts and prices

The second workflow can reduce retrieval effort and make answers easier to reuse. But a generated answer is not automatically more accurate or more authoritative than the underlying website. Agents should still verify important claims against primary product pages, current pricing and availability.

Also, being discoverable is not the same as being actionable. The presence of an MCP or API endpoint advertises a potential integration; it does not prove that an agent can successfully book a rental, buy a phone, make a payment or complete another transaction.

4. Verdict

No-BAES is not AI-invisible. Dronajmij already provides a useful discovery index with meaningful business and product information. Conventional crawling can still produce answers and citations.

BAES has a demonstrated advantage in answer retrieval. MentalQuest's BCP directory and live question endpoint give an agent a more direct path from a question to a structured, relevant business answer.

This five-URL test does not prove that every BAES-enabled website outperforms every No-BAES website. Only MentalQuest's BCP flow was successfully verified end to end. The other BAES-hosted endpoints need successful requests and equivalent test questions before their actual answer quality can be compared.

The strongest conclusion is that BAES can improve the path from discovery to a focused, citable answer. To establish a measurable readiness advantage, the next step is to run identical questions against both groups and score answer accuracy, source attribution, price completeness, freshness and successful task completion.

Live test results

I checked all six URL variants you listed (the two MentalQuest question variants, the OpenAPI document, the BCP pages index, and the two Dronajmij index URLs). The most interesting result is that the two MentalQuest query parameters return meaningfully different answers to the same question, while the raw Dronajmij index succeeds and its hyphenated preview returns a gateway error. 
/bcp/ask Answer Engine
+3

1. MentalQuest: baes-rc vs. develop
① tracy_infra=baes-rc

Open live answer 

The endpoint returns a concise, integrated recommendation:

Archetype: Dionysus, linked to sensory aliveness, emotional openness and creativity, with anxiety and sensory overload described as shadow aspects.

Inner-child program: Inner Child Contact, about 30 minutes, 410 UAH.

Immediate anxiety support: Escaping Chaos, 18 minutes, 99 UAH.

It connects the two needs and asks whether the user wants links to either technique. 
/bcp/ask Answer Engine

② tracy_infra=develop

Open live answer 

This version gives a more explanatory, structured answer:

Anxiety: Dionysus, with a link to the archetype assessment.

Inner-child work: Inner Child Contact, around 30 minutes, 410 UAH, plus mention of a companion workbook.

Gentler starting point: Escaping Chaos, 18 minutes, 99 UAH.

It explicitly separates the archetype recommendation from the therapeutic technique and then explains how the two can work together. 
/bcp/ask Answer Engine

What changed?

Dimension

	

baes-rc

	

develop




Main answer

	

Integrated recommendation

	

Step-by-step explanation




Archetype

	

Dionysus

	

Dionysus




Inner-child price

	

410 UAH

	

410 UAH




Anxiety-support price

	

99 UAH

	

99 UAH




Additional details

	

Emphasizes emotional safety and stabilization

	

Adds assessment link and companion workbook




User experience

	

Shorter, more conversational

	

More explicit reasoning and separation of needs

Interpretation: Both variants agree on the core recommendation and prices, but they do not return identical content. This demonstrates variation between the two query modes or serving configurations, not necessarily a quality regression or improvement. A single request to each URL cannot establish whether the difference is deterministic, caused by different prompts, or due to changing backend content.

2. OpenAPI: machine-readable API contract

③ MentalQuest /openapi.json 

HTTP content retrieved

The endpoint returns an OpenAPI 3.1.0 document for the MentalQuest BAES website API, version 2.0.0. The document describes the service as offering a free archetype test, a 99 UAH trial technique and specialized techniques at 410 UAH.

The retrieved specification advertises these routes:

Endpoint

	

Declared purpose




GET /bcp/ask

	

Ask the AI concierge




POST /bcp/mcp

	

MCP JSON-RPC endpoint




GET /bcp/api/query

	

Query business records and programs




GET /bcp/api/forms

	

List structured interaction and booking forms




GET /bcp/api/menu

	

Retrieve the service and program catalog




GET /llms.txt

	

Retrieve AI-oriented website context

mentalquest.app

AI-readiness implication: This is more useful to an API-capable agent than an ordinary page index because it advertises machine-readable operations and their intended purposes. However, the successful retrieval of the specification does not prove that every declared route works. The fetched response was also rendered as a compact JSON document, so this test establishes that the contract is discoverable, not that every schema, parameter, response type or operation has been validated end to end.

3. BCP pages: discoverable, product-specific landing pages

④ MentalQuest /bcp/pages 

Page retrieved

The page documents four canonical landing-page examples, each connected to an underlying WordPress post. It also lists different presentation modes and assistant personas. 
mentalquest.app

The four showcased offerings are:

Escaping Chaos — 99 UAH.

Animus Bond — 410 UAH.

Eco-Aggression Meditation — 410 UAH.

Balance Yin & Yang — 410 UAH.

The page exposes alternative assistant configurations (default, somatic-emergency, jungian-shadow and cbt-neuro-coach) and presentation modes such as inline, floating, split and conversational.

AI-readiness implication: This is a useful catalog of landing pages and their intended user experiences. It gives an agent more than a bare list of URLs: it connects canonical content to specific programs, prices and specialized interaction modes. It is primarily a showcase and discovery page, though—not proof that each configuration or interaction mode works correctly in a live browser.

4. Dronajmij: hyphenated preview vs. raw website
⑤a. Hyphenated BAES preview

dronajmij-pl.baes.tracy.io/llms.txt 

502 Bad Gateway

The request failed with a gateway error. No index content could be inspected, so this test cannot establish what the preview would return if it were healthy. 
Internal Error

⑤b. Raw production index

dronajmij.pl/llms.txt 

Content retrieved

The raw endpoint returns a 124-line index generated by Yoast SEO. It links to service pages, articles, product listings, product categories, brands and the sitemap. 
dronajmij.pl

The raw index includes concrete rental details:

DJI Neo 2: refundable PLN 200 deposit; daily rental price depends on the number of days.

DJI Mini 5 Pro: refundable PLN 400 deposit; daily rental price depends on the number of days.

Starlink rentals: refundable PLN 300 deposit and a seven-day minimum rental period.

The product descriptions also contain equipment inclusions and links to contact the business. One limitation is that some HTML markup is escaped inside the index, which makes the content less clean for downstream extraction. 
dronajmij.pl

Important distinction: The hyphenated host is a separate preview hostname, not simply another spelling of the production URL. A 502 on that host says something about the availability of that particular route at test time; it does not show that the raw website's index is broken or that the BAES implementation itself is inherently incapable of serving it.

Final comparison

URL

	

What it returned

	

Main value for AI agents




① MentalQuest baes-rc

	

Concise answer with archetype, two programs and prices

	

Fast, focused retrieval




② MentalQuest develop

	

More detailed answer with assessment link and workbook

	

More context for decision-making




③ MentalQuest OpenAPI

	

OpenAPI 3.1.0 contract

	

Machine-readable API discovery




④ MentalQuest BCP pages

	

Four canonical program pages and presentation configurations

	

Product discovery and specialized landing pages




⑤a Dronajmij preview

	

502 gateway error

	

No usable evidence from this request




⑤b Dronajmij raw index

	

SEO-generated llms.txt

	

Broad content discovery and some product facts

Bottom line

MentalQuest demonstrates a richer observable AI interface in this test: a question-answer endpoint, a published API contract and a landing-page catalog with product-specific information. Its two answer variants also agree on the main recommendation and prices despite different wording.

Dronajmij's production llms.txt remains useful for discovery, but it is less convenient for direct question answering and contains escaped markup. Its BAES preview was unavailable during this test, so the evidence supports a comparison of the working production index against the working MentalQuest interfaces, not a conclusion that the preview's intended functionality is inferior.

The next meaningful test would be to query the documented catalog and MCP endpoints directly, verify their response schemas, and compare the same product questions against both sites for factual accuracy, citation quality, price completeness and response consistency.

Live URL comparison: BAES-enabled vs. No-BAES

The six endpoints reveal an important distinction: a URL can exist and return useful data while still being unreadable to a particular inspection tool. MentalQuest's configuration and verification APIs return structured JSON; its JSON-LD and full-text resources are rejected by the inspection tool because of their content types. The two No-BAES llms.txt checks do not yield readable index content in this test. 
Internal Error
+5

1. What each URL returns

① MentalQuest — JSON-LD schema

Open schema.jsonld 

Content not parsed

The inspection tool reports HTTP-style error 400 Unsupported content-type: application/ld+json. This is a parser/content-type limitation, not proof that the endpoint itself returns an HTTP 400 to normal clients. The actual JSON-LD payload could not be inspected, so its schema types and business facts remain unverified. 
Internal Error

② MentalQuest — BCP configuration

Open /bcp/config 

JSON retrieved

Returns a JSON configuration object with status: ok, service name bcp-baes, gateway version 1.0.0-rc.6, and the active target baes-rc. It lists three supported infrastructure targets—baes-rc, develop, and production—as well as fallback behavior and endpoint configuration.

AI value: Strong operational discoverability. An integration can learn which environments are advertised and how the service is configured. However, this is infrastructure metadata, not a catalog of customer-facing offerings. 
mentalquest.app

③ MentalQuest — domain verification

Open /bcp/verify 

JSON retrieved

Returns status: verified and confirms ownership verification for mentalquest.app. But the same response reports that the root domain's CNAME verification and SSL status are still pending, while the www hostname and its SSL status are active.

AI value: Useful for diagnosing deployment and domain-routing readiness, but not directly useful for answering customers' questions about programs or prices. The endpoint also exposes detailed DNS and hosting configuration, which should be reviewed for unnecessary public disclosure. 
mentalquest.app

④ MentalQuest — full LLM text

Open /llms-full.txt 

Content not parsed

The inspection tool reports 400 Unsupported content-type: text/markdown. It did not expose the text body, so the contents, length and completeness of the full-text index cannot be confirmed from this request.

AI value: Potentially high if the endpoint supplies consolidated, accurate business information, but this particular test does not establish what the resource contains. 
Internal Error

⑤ Books to Scrape — llms.txt

Open books.toscrape.com/llms.txt 

404 Not Found

The requested path returns a 404 in the inspection result. No llms.txt index was available to read at that URL.

AI value: No benefit from this specific discovery file. That does not mean the book catalog itself is inaccessible to conventional crawlers or that other structured data is absent. 
Internal Error

⑥ Yes Plz Coffee — llms.txt

Open yesplz.coffee/llms.txt 

Redirect target inaccessible

The request redirects to www.yesplz.coffee/llms.txt, but the inspection tool cannot access that destination. It reports a 500-style fetch failure associated with the inaccessible redirect target.

AI value: Inconclusive. Unlike the Books to Scrape result, this does not establish that the index is missing; the redirected resource simply could not be inspected. 
Internal Error

2. Comparison by AI-readiness capability

Capability

	

MentalQuest

	

Books to Scrape

	

Yes Plz Coffee




Machine-readable configuration

	

Verified JSON endpoint

	

Not tested

	

Not tested




Domain verification diagnostics

	

Verified JSON endpoint

	

Not tested

	

Not tested




Structured semantic metadata

	

JSON-LD endpoint exists, payload not inspected

	

Not tested

	

Not tested




Full-text AI index

	

Endpoint exists, body not inspected

	

Requested index returned 404

	

Redirect target inaccessible




Customer-facing AI answers

	

Demonstrated in the earlier /bcp/ask test

	

Not tested

	

Not tested

The most important qualification is that this is not an apples-to-apples content test. MentalQuest's configuration and verification endpoints are specialized APIs, while the comparison URLs for the other two sites are only llms.txt files. Their absence or inaccessibility does not prove that their websites have no structured data, product metadata or other AI-accessible content.

3. What the results mean

MentalQuest exposes a broader integration surface. Its configuration and verification endpoints provide explicit machine-readable operational information. Together with its previously tested BCP question-answer endpoint, this makes the service easier to interrogate programmatically.

Successful HTTP retrieval is not the same as successful AI consumption. The JSON-LD and Markdown resources were rejected by the inspection tool's content-type handling. A robust AI-readiness audit should separately test whether the resource exists, whether its content type is correct, and whether a client can parse its payload.

No-BAES sites can still be AI-readable. A missing llms.txt does not prevent an agent from crawling ordinary pages, following product links or using other structured metadata. It merely removes one potential discovery mechanism.

Recommended next checks

Fetch the JSON-LD and full-text endpoints with a client that accepts their declared content types, then validate the JSON-LD syntax and inspect the actual Markdown body.

Test the redirected www.yesplz.coffee/llms.txt URL directly and check whether Books to Scrape exposes any alternative machine-readable catalog or sitemap.

Compare all three businesses using the same questions and score the answers for factual accuracy, product coverage, price completeness, citation quality and successful retrieval.

Bottom line: MentalQuest demonstrably provides machine-readable configuration and domain-verification endpoints, but two of its content endpoints remain unverified in this inspection. The No-BAES results show unavailable discovery files—not necessarily inferior underlying content or a definitive lack of AI readiness.

Results at a glance

All six URLs were tested on 23 October 2026. Four returned readable content; two returned 404 errors. The most useful finding is that MentalQuest exposes not just an AI answer service, but a structured catalog, schema definitions and product-specific pages.

#

	

Endpoint

	

What it returned

	

Practical value




1

	

/bcp/api/query 

	

JSON business records

	

High — products, prices, durations, target needs, assessments and consultations




2

	

/bcp/api/forms 

	

JSON form and field definitions

	

High — explains data structures and relationships between products, assessments, bookings and inquiries




3

	

/bcp/api/menu 

	

JSON catalog organized into sections

	

High — simple, readable offering and price list




4

	

Balance Yin & Yang page 

	

HTML product page rendered in a chat-oriented wrapper

	

High — explanatory content and purchasing form




5

	

Archetypes test page 

	

404 Not Found

	

Unavailable at the exact URL tested




6

	

BAES GetVersionInfo 

	

404 Not Found

	

Unavailable at the exact URL tested

Sources: 
mentalquest.app
+5

1. /bcp/api/query: the richest data endpoint

The JSON response contains seven therapeutic techniques, four archetype assessment profiles, and consultation records. It provides machine-readable facts that are difficult to extract reliably from marketing prose alone. 
mentalquest.app
+1

Examples of returned data:

Escaping Chaos: 99 UAH, 18 minutes, targeting acute anxiety and sensory overload.

Inner Child Contact: 410 UAH, 30 minutes, targeting emotional numbness, fear of abandonment and perfectionism.

Gender Energy Balance: 410 UAH, 38 minutes.

Archetype profiles for Ares, Athena, Artemis and Dionysus, each listed at 0 UAH.

Consultations with listed prices of 850, 950, 1,200 and 1,500 UAH across different formats and programs.

One noteworthy data-quality issue: some records appear in both a techniques array and a therapeutic_techniques array, with slightly different field types—for example, prices are numeric in one representation and strings in another. Agents should normalize these before using the data in calculations.

2. /bcp/api/forms: how the data is organized

The response defines fields for therapeutic techniques, archetype assessments, somatic consultations and client inquiries. It includes field types, required-field flags, and links between records—for example, an archetype assessment can reference a recommended therapeutic technique. It also exposes form definitions and printable-document templates. 
mentalquest.app
+1

This is useful for agents and integrations that need to understand not only what a product is, but how it relates to a booking or inquiry workflow.

A caution: the endpoint exposes fields for client names, email addresses and phone numbers. These are field definitions, not evidence that customer records were returned. Still, public access to internal form structure should be intentional.

3. /bcp/api/menu: the simplest catalog for AI

This is the easiest of the three APIs for a general-purpose AI agent to consume. The catalog is grouped into introductory practices, therapeutic deep-dives, and interactive tools/community programs. It lists prices and canonical URLs directly. 
mentalquest.app

It includes:

Escaping Chaos — 99 UAH.

Five therapeutic techniques plus Archetype Activation Meditation — 410 UAH each.

Archetype Discovery Test — free.

Community membership and festivals — admission is listed without a fixed numeric price.

Compared with /api/query, this menu has less clinical or psychological detail but is easier to turn into a short recommendation or price comparison.

4. Balance Yin & Yang: real content and a purchase path

The product-specific page returned readable HTML in Ukrainian. Its heading identifies the technique as Balance Yin & Yang, priced at 410 UAH, and the body explains its intended purpose and use. It includes preparation instructions, guided-practice steps, follow-up advice, related techniques and a purchase form with required phone and email fields. 
MentalQuest

This complements the JSON endpoints: the APIs supply structured facts, while the page supplies explanatory context and a customer-facing path toward purchase.

5–6. Two unavailable URLs

The exact archetypes-test page under /bcp/pages/blog/archetypes-test/ returned 404. The API menu nevertheless points to a separate working-looking canonical path, https://mentalquest.app/discover-your-archetype. That is a useful alternative to test, but it is not proof that the requested BCP page works.

The development version-information endpoint also returned 404. This only establishes that the specified route was unavailable to the inspection tool; it does not prove that the entire development API is down or that version information is not exposed elsewhere.

Overall assessment

MentalQuest's strongest advantage is the combination of three complementary layers:

Structured facts: /bcp/api/query supplies detailed records.

Discoverable relationships: /bcp/api/forms documents schemas and references.

Simple product discovery: /bcp/api/menu supplies a clean catalog, while product pages provide human-readable context.

That combination is particularly useful for AI agents that need to answer questions about offerings, compare prices, recommend a relevant product and link users to the source. The two 404s show that endpoint availability and URL consistency still need attention, and the duplicated data structures mean that agents should verify prices and canonical URLs before taking consequential actions.

Попрацював 20s
Final AI-readiness comparison: BAES vs. No-BAES

The latest tests show that BAES-enabled websites can expose live, question-oriented business answers, while conventional websites can also provide excellent machine-readable instructions for AI agents. The strongest example of the latter is iFixit's llms.txt, which explicitly directs agents to a structured API, explains citation requirements and links to an MCP integration. 
Certified Refurbished Phones & Repairs
+1

1. Latest five URL checks

Endpoint

	

What the inspection returned

	

Assessment




1. PhoneFixer /bcp/ask

	

A direct answer with an iPhone battery-replacement price, refurbished-device inventory, battery health, grades and warranties

	

Strong live business-answer capability




2. iFixit /llms.txt

	

A 102-line text guide describing its API, canonical pages, attribution rules, licensing and MCP endpoint

	

Excellent AI discovery and usage guidance




3. MentalQuest homepage

	

A minimal rendered concierge label: “Standard Concierge · Ask Mentalquest”

	

Widget is discoverable; Schema.org markup and direct /bcp/ask links were not verified




4. Two readiness /health URLs

	

The BAES hostname rendered a concierge label; the non-BAES hostname returned no readable lines

	

Inconclusive health-check comparison




5. Castorama /llms.txt

	

Redirected to www.castorama.pl/llms.txt, which the inspection tool could not access

	

Index availability unverified, not proof that the file is absent

Sources: 
Certified Refurbished Phones & Repairs
+5

PhoneFixer: the clearest live business example
Answer returned with inventory checked on 23 October 2026

iPhone 13 battery replacement: €45, including parts and labour, approximately 45 minutes, with a six-month warranty.

iPhone 13 128GB Midnight, Grade A, 92% battery health: €239, with a 12-month warranty.

iPhone 13 Pro 256GB Graphite, Grade A, 94% battery health: €349, with a 12-month warranty.

The answer also lists an iPhone 14 at €259 and a Samsung Galaxy S23 at €289.

Source: Live PhoneFixer answer . These are the values returned by the endpoint, not an independent audit of physical stock.

This is a particularly useful demonstration because the question combines two different needs—repair pricing and product availability—and the endpoint returns both in one answer, with a verification timestamp.

2. What the cumulative experiments demonstrate

Across the URL checks in this conversation, the evidence falls into four categories.

A. Live answers and business context

Strongest demonstrated capability: MentalQuest and PhoneFixer.

Their working /bcp/ask endpoints answer natural-language questions with specific offerings, prices and relevant context. PhoneFixer additionally returns a time-stamped inventory answer. MentalQuest's tests demonstrate that different infrastructure targets can return differently structured responses to the same question.

B. Structured catalogs and integration interfaces

Strongest demonstrated capability: MentalQuest's API family.

The earlier tests retrieved the BCP query, forms and menu APIs, the OpenAPI contract, configuration and verification endpoints, and product-specific content. Together, these give agents multiple ways to discover offerings, understand data relationships and retrieve customer-facing information.

The caveat is that not every endpoint worked: some URLs returned 404s, and the JSON-LD and Markdown resources were rejected by the inspection tool's content-type handling.

C. Conventional AI discovery can be excellent

Strongest demonstrated capability: iFixit and Dronajmij's raw index.

iFixit's llms.txt is unusually explicit about how agents should retrieve information, cite sources, respect licensing and use structured API responses rather than scraping HTML. Dronajmij's production index also exposes useful service and product links and some rental details.

This is an important counterexample to the idea that a website needs BAES to be AI-readable: a well-designed conventional index and API can already provide strong agent access.

D. Endpoint availability is a separate readiness dimension

The tests encountered missing paths at Chleboteka, Books to Scrape and some MentalQuest endpoints, a gateway error on the hyphenated Dronajmij preview, and inaccessible redirects for Yes Plz Coffee and Castorama. The latest readiness health URLs were also inconclusive.

These outcomes identify routes that need further verification; they do not prove that the entire website or integration is unavailable.

3. What about Schema.org QAPage, SearchAction and health checks?

These need more precise verification before drawing conclusions.

MentalQuest homepage: the rendered result exposes a concierge label, but this test did not retrieve the raw HTML or structured-data graph. I therefore cannot confirm whether QAPage, SearchAction, or a direct /bcp/ask link is present in the page source.

Readiness endpoints: the two URLs did not return comparable, readable health-status payloads through this inspection. A proper comparison needs the raw HTTP status, content type and response body from each hostname.

Citation quality: an endpoint returning an answer is not sufficient on its own. The best implementations should let agents trace important facts to canonical product pages or source records, particularly when prices and inventory change.

4. Overall scorecard

This is a qualitative assessment of the observed evidence, not an official readiness score.

Dimension

	

BAES-enabled examples

	

No-BAES examples




Natural-language business answers

	

Strong where /bcp/ask works

	

Not tested equivalently for all sites




Structured business records

	

Strong in MentalQuest's APIs

	

iFixit offers a well-documented public API




Discovery documentation

	

BCP directories and API contracts

	

iFixit's explicit llms.txt; Dronajmij's index




Live inventory and price answers

	

Demonstrated by PhoneFixer

	

Requires a comparable live query to assess




Citation and usage guidance

	

Varies by endpoint

	

Particularly explicit in iFixit's index




Reliability of every tested URL

	

Mixed

	

Mixed; some indexes are missing or inaccessible

Final conclusion

The experiments support a narrower but more defensible conclusion than “BAES always wins”:

BAES can make business knowledge more directly queryable and actionable for AI agents. MentalQuest demonstrates this through a combination of question answering, structured catalogs and integration endpoints; PhoneFixer demonstrates it through live repair pricing and refurbished-device inventory.

But BAES is not a prerequisite for high AI readiness. iFixit's conventional llms.txt provides an excellent roadmap to structured content, API usage, citations and MCP. Meanwhile, several BAES-related preview or auxiliary endpoints failed to return inspectable content.

The most meaningful benchmark is therefore not whether a website has BAES, but whether an agent can reliably discover the right information, return accurate and current answers, cite the underlying source, and complete the intended task. A fair final benchmark would run the same questions against each site, validate responses against canonical records, and measure retrieval success, factual accuracy, citation coverage, freshness and endpoint reliability.

ChatGPT — це ШІ й може помилятися.