News APIs need audit trails, not chatbot summaries

The obvious use for AI in news is summarization. That is also the boring one.

A summary helps for about thirty seconds. Then the harder questions arrive. Who said this first? Is the source primary or recycled? When did the system first see it? Did the article change? Can another agent cite the same claim without laundering it through a paragraph of generated mush?

That is why I keep coming back to Asha.News as a product problem, not a content problem. The homepage promises "News your AI can cite. Sources you can check." The agent API page backs that up with public read endpoints, /api/v1/feed, /api/v1/search, /api/v1/digest, article share endpoints, story share packs, source health, and an OpenAPI contract. The current public feed also returns timestamps such as source_published_at, asha_first_seen_at, asha_last_checked_at, and freshness_status for each item.

Sources: Asha.News homepage, Asha.News Agent API wiki, Asha.News public feed, Asha.News public digest.

The timestamp is part of the claim

Most AI news products treat time as metadata. It sits under the card, smaller than the headline, as if it were decoration.

For agents, time is load-bearing.

If a system tells me a story is fresh, I want to know which clock produced that judgment. Asha.News separates the source publication time from the moment its own crawler first saw the item and the moment it last checked it. That sounds like plumbing because it is plumbing. Good plumbing is underrated.

A story can be published at 04:01, seen by Asha.News at 04:20, and checked again at 04:20. Those are different facts. Collapsing them into "today" destroys the part an agent needs for verification.

The product rule is simple: never make the model infer freshness from prose. Put the clocks in the API.

Evidence should travel with the excerpt

A daily digest is useful when I am browsing. It is dangerous when an agent treats it as the source of truth.

The better pattern is a digest that points back to evidence. Asha.News already exposes article pages, article share endpoints, story share packs, source health, and public search. That gives an agent a path from "this looks relevant" to "here is the specific article/story URL I can cite" to "here are the underlying sources I should inspect before I make a factual claim."

That path matters because generated summaries blur authorship. A paragraph can be accurate and still be a bad citation if nobody can tell where the sentence came from. The citation trail should be boringly mechanical:

claim -> Asha.News story or article -> original source -> timestamp -> confidence/evidence notes

If one link is missing, the agent should say so. Better an awkward citation than a confident invented one.

The API is the interface

Readers see cards. Agents see contracts.

That is the design shift I like here. Asha.News is not only presenting a news feed to humans. It is exposing a versioned /api/v1 surface, public MCP over JSON-RPC 2.0, no-key public reads, OpenAPI discovery, and endpoint groups for feeds, search, sources, Palestine desk data, conflict monitor events, markets, and share packs.

That is not glamorous. It is exactly the kind of unglamorous decision that makes a product useful inside an automation loop.

A browser card can hide ambiguity with layout. An API cannot. If the field is missing, stale, badly named, or mixed with editorial copy, the next agent will either guess or fail. Guessing is how a news assistant becomes a rumor compressor.

My bar for agent-readable news

I do not want an AI news product that writes smoother recaps. I want one that makes bad citations harder.

That means:

• Source URLs stay visible.
• System timestamps stay separate from publisher timestamps.
• Search returns stable article or story URLs, not just snippets.
• Digest text is treated as a starting point, not the final citation.
• Public endpoints are documented well enough that another agent can repeat the lookup.

Asha.News is closest to the right shape when it behaves less like a newsletter and more like an evidence router. The headline gets my attention. The API earns my trust.

That is the part most AI news products still miss. The product is not the summary. The product is the chain of custody behind the summary.