A comparative look at the BCCL–OpenAI partnership and the "Citing Agent" architecture
The trigger
This week's news: Bennett, Coleman & Co Ltd (BCCL) — publisher of The Times of
India, The Economic Times, and a stable of other titles — announced a strategic
partnership with OpenAI to make its journalism more discoverable inside ChatGPT.
Under the arrangement, when ChatGPT surfaces a summary or excerpt from a
BCCL publication, the response will name the publication and link back to the
original article. BCCL retains full editorial control; OpenAI has no say in what gets
reported or how.
It's a good deal, and a sensible one for a media house of BCCL's size. But it is
worth being precise about what kind of solution it actually is — because it is easy
to read a headline like this and assume the "AI discoverability" problem has one
shape, and one price tag: a negotiated partnership with a large AI lab.
It doesn't. And I want to lay out why, using the architecture I've been building
with my technical collaborator Kishan — what I've been calling the Citing Agent
— as the contrast case.
To be clear about what this post is and isn't : it isn't a claim that BCCL copied
anything, borrowed anything, or was even aware anything like the Citing Agent
exists.
The two efforts were conceived independently, for different scales of
publisher, solving the same underlying shift — AI answer engines becoming a
discovery layer for content — in two structurally different ways.
That contrast is the actual subject of this post.
Two ways to get cited by an AI
BCCL's route is a bilateral commercial partnership. OpenAI and BCCL sat
down, negotiated terms, and built a pipe between BCCL's content and ChatGPT's
responses. It is a licensing-and-integration relationship, and — like all such
relationships — it is scoped to the two parties who signed it. It gets BCCL
surfaced in ChatGPT. It says nothing about Perplexity, Gemini, or Claude, unless
BCCL negotiates separate deals with each of them.
The Citing Agent's route is different in kind. It doesn't ask any AI company for
permission. It makes a blog corpus — in this case, mine: 6,800+ posts going back
to 2002 — discoverable and citable using open, standard mechanisms that any
AI answer engine can choose to respect :
> structured schema markup,
> an llms.txt manifest,
> proper crawler access, and (going further than passive discoverability)
> a live, queryable tool that an AI agent can call directly.
One is a deal. The other is infrastructure.
Side-by-side
| Dimension | BCCL–OpenAI Partnership | The Citing Agent |
|---|---|---|
| Core mechanism | Bilateral commercial deal — OpenAI gets negotiated access to surface BCCL content inside ChatGPT | Self-service technical pipeline — no deal with any AI vendor; works via open web standards |
| Who has to say yes | OpenAI (and, separately, any other AI vendor BCCL wants coverage from) | Nobody — respects standards that any crawler or AI agent can choose to use |
| Scale of applicability | One publisher, one AI engine (ChatGPT), per signed deal | Any AI engine that respects schema/llms.txt/crawling conventions — designed with ChatGPT, Perplexity, Gemini, and Claude all in view |
| Content corpus | Live, continuously produced newsroom reporting | A static, growing archive — framed as a timestamped intellectual prior-art trail |
| Technical layer | Undisclosed — presumably a backend licensing/API integration | Fully specified 8-stage pipeline: blog vectorization, news scanning, semantic matching, draft generation, infographic generation, an approval gate, and publish/send |
| Retrieval style | Passive — OpenAI decides if/when/how BCCL content appears in a response | Two parallel roads — a passive AEO road (structured pages any crawler can index) and an active MCP road (a live tool, SearchMyBlogs, an AI agent can query directly) |
| Attribution mechanism | Contractual — ChatGPT is obligated by the deal's terms to name the publication and link the article | Structural — entity/Person schema and direct URLs are baked into how the content is served, so attribution isn't a negotiated favor, it's how the pipes are built |
| Business model | Commercial partnership (terms undisclosed) | No commercial layer at all — the goal is discoverability and citation credit, not monetized access |
| Replicable by a small/independent publisher? | No — needs the negotiating leverage of a large media house | Yes — this is precisely the point: it's what a blogger, a startup, or a small publication can do without that leverage |
----------------------------------------------------------------------------------
The technologies behind the Citing Agent, and why each one is there
It's worth walking through the stack, because none of the choices are arbitrary —
each one answers a specific problem in getting an old, large, unglamorous blog
archive taken seriously by machine readers.
1. An llms.txt manifest.
This is the emerging convention for telling an AI crawler "here is what this site is,
and here is where the good stuff lives" — a machine-readable table of contents,
distinct from a human-facing sitemap. Without it, an AI engine crawling 6,800
posts has no signal about which ones matter, which are duplicates, or how they
cluster into themes. It's the cheapest, highest-leverage piece of the whole
architecture.
2. Schema markup and an entity layer.
Structured data (schema.org-style markup, FAQ blocks, and — the more recent
addition — a Person/entity schema) turns a blog post from "some text on a page"
into a machine-parseable claim: this specific person said this specific thing on this
specific date. The entity layer was added after noticing a Google patent on entity-
level LLM understanding — the insight being that being indexed and being cited
are not the same achievement. An AI engine can crawl a page without ever
attaching it to a named, verifiable author. The entity layer exists to close that gap.
3. IndexNow and the Bing URL Submission API.
These are the plumbing that gets new or updated content in front of search and AI
crawlers quickly, rather than waiting for a slow, undirected crawl cycle. One
wrinkle worth mentioning: because the archive lives partly on a Blogger/blogspot
subdomain (Google-owned), a root-level IndexNow key file can't be hosted there
— so Bing's own URL Submission API was integrated directly instead, as a
workaround specific to that hosting quirk.
4. Vector search — SearchMyBlogs.
This is the "active road." Rather than waiting passively to be crawled, it's a live,
queryable retrieval tool — built on Kishan's Brihas infrastructure — that an AI
agent can call directly via MCP (Model Context Protocol) to search the blog corpus
semantically, not just by keyword. This is the piece that turns the archive from "a
website that might get crawled" into "a tool an AI agent can actively consult,"
which is a meaningfully stronger form of citability than indexing alone.
5. The 8-stage autonomous pipeline.
Beyond static discoverability, the fuller design (vectorization → news scanning →
semantic matching → draft generation → infographic generation → an approval
gate → publish/send) is built to actively match archive posts against current news
events — surfacing decades-old writing as timely commentary, with a human
approval step before anything goes out. This is what makes a 2002 blog post
about, say, education policy, relevant to a 2026 headline, instead of sitting inert in
an archive.
Why this matters beyond one blog
I have zero commercial interest in this comparison — I'm not selling a product,
and I'm not trying to compete with BCCL or OpenAI. If any startup, independent
publisher, or fellow blogger finds the architecture above useful and wants to build
their own version, they're welcome to it. The whole point of laying it out this
plainly is that the "AI discoverability problem" doesn't have to be solved the
expensive way.
BCCL's deal proves the discoverability problem is real and worth solving.
The Citing Agent is offered as evidence that it doesn't require a seat at
OpenAI's negotiating table to solve it.
============================================
with Regards,
Hemen Parekh
www.Iam-Immortal.ai
Source:
BCCL partners with OpenAI to enhance journalism discovery for readers globally, The Economic Times.
No comments:
Post a Comment