How late is your market news, really?
We polled 13 public financial news feeds every 20 seconds and wrote down the exact second each headline became visible. Window: 2026-08-13 18:28 UTC → 2026-08-13 18:44 UTC (16.3 minutes), 298 headlines captured, 46 of them published while we watched. Method, raw numbers and CSV below — check the work.
Read the caveats first
Everything below is one observation window on one day, and the metric has real limits. State them up front or the numbers mean nothing:
- This measures feed visibility, not newsroom speed. The clock starts at the timestamp the publisher itself put on the story and stops when the item is fetchable from the public endpoint. A newsroom can be fast and its RSS pipeline slow.
- Publisher timestamps are self-reported. Some round to the minute, some stamp the story at write time rather than release time. That noise is baked into every number here.
- Polling every 20s puts a floor under the measurement. On average we discover an item ~10s after it truly appears, so read small differences as noise.
- A feed’s first fetch returns its whole backlog, which is not evidence of anything. Those items are excluded from every latency figure; only headlines that appeared during the run are counted (36 of them were genuinely fresh and timestamped). Items re-emitted with a timestamp already hours old are excluded from the latency numbers too, and counted separately — that turned out to be its own finding.
- 16.3 minutes is a short window. It is enough to size the delay in orders of magnitude and to show the overlap effect. It is not enough to crown a winner.
Publish → visible delay, by source
Lower is better. “Fastest” is the single quickest headline we caught from that source — the best case the feed is capable of, versus the median it actually delivers.
| Source | Median | p90 | Fastest | New items | Items/hr |
|---|---|---|---|---|---|
| GlobeNewswire (PR) | 1.9 min | — | 1.9 min | 1 | 3.7 |
| Seeking Alpha | 2.2 min | 4.2 min | 38s | 12 | 44.3 |
| PR Newswire | 2.2 min | 2.3 min | 75s | 9 | 33.2 |
| Investing.com | 7.3 min | 8.4 min | 2.9 min | 6 | 22.2 |
| Reuters (via Google) | 16.0 min | 35.4 min | 88s | 8 | 29.5 |
Feeds that published nothing new during the window (so no lag could be measured), with the size of the backlog we captured: Benzinga (10), Business Wire (23), CNBC Finance (30), CoinDesk (25), Dow Jones / WSJ Markets (20), Federal Reserve (20), MarketWatch Top Stories (10), Yahoo Finance (53).
The finding that surprised us: the feeds barely overlap
The intuition is that all the majors carry the same wire copy, so watching one or two is enough. In this window that was not true. We clustered the 46 new headlines into 45 distinct stories by title similarity — 45 of them (100%) were carried by exactly one feed, and only 0 appeared on two or more inside the window.
Some of that is genuine editorial difference: a press-release wire, a crypto desk and a central-bank feed simply do not cover the same events. Some of it is timing — a slower feed may have carried the same story after we stopped watching. Both readings point the same way: the feed you are watching is showing you a slice, and the size of the slice is smaller than it feels.
In this window no single story was picked up by two different feeds inside the observation period, which is itself the finding below: these feeds barely overlap in real time.
The other kind of stale: archive content in the fresh slot
Latency assumes the item is news. Some feeds also rotate old articles into the feed as if they were new. We counted every headline that appeared during the window carrying a publisher timestamp already more than six hours old: 10 of 46 new items (21.7%) — Yahoo Finance (10 items, median stated age 34.1 hours).
These are excluded from the latency table, because calling a 34.1-hour-old republish a “34.1-hour delay” would be nonsense. But they matter in their own right: if you are eyeballing a feed for something new, a slice of what you see is not new at all, and nothing in the feed marks the difference.
What this means if you actually trade the news
Two separate problems fall out of the numbers, and they need different fixes.
- Latency. A median of 2.7 min and a slow decile of 13.2 min is an eternity against an algorithmic reaction measured in milliseconds. If your entry depends on being early, a public RSS poll is not a news source, it is a confirmation source.
- Coverage. Even a fast feed only shows you its own desk. At 100% single-source stories, adding feeds buys you more marginal information than upgrading any one of them — up to the point where you cannot read them all, which is the real reason people consolidate.
That consolidation problem is what Top Tier Newswire is built for: 40+ sources merged into a single tape with sentiment tagging and per-ticker watchlists, quoted at sub-two-second latency, with a free delayed tier you can look at before deciding whether the real-time tier is worth it. We think it is the right shape of answer to the coverage number above; the latency claim is theirs and we have not independently benchmarked their pipeline, so measure it against your own feed before you rely on it.
Disclosure: Smeltworks has a commercial relationship with Top Tier Newswire. The benchmark above was run against public feeds and does not include their product; every number on this page is reproducible from the method below.
Method — reproduce this yourself
- Take the 13 public RSS/Atom endpoints listed in the CSV.
- Fetch every one on a 20-second loop. Key each item by source + guid/link + title.
- The first time a key is seen, record wall-clock
first_seenalongside thepubDate/publishedthe publisher supplied. - Discard each source’s first successful poll (backlog, not news). Discard items whose stated time is more than 6 hours old or more than 2 minutes in the future — those are re-publishes and clock skew, not latency.
- Latency =
first_seen − pubDate. Report the median, not the mean: a handful of stale re-posts drags a mean anywhere you like. - For overlap, normalise titles to lowercase token sets, drop stopwords, and cluster at Jaccard ≥ 0.5. Count how many clusters contain more than one source.
Run on 2026-08-13 18:45 UTC. 32 fetch errors were recorded and excluded. Feeds that returned nothing new are shown separately rather than dropped, because a quiet feed is a real result too.
Questions
What does “publish → visible” latency mean?
It is the gap between the timestamp a publisher puts on a story and the moment that story actually shows up in its public feed, where a reader or a script can see it. It is the delay you personally experience, not the publisher’s internal one.
Which financial news feed is fastest?
In this run, GlobeNewswire (PR) had the lowest median publish-to-visible delay at 1.9 min, and Reuters (via Google) the highest at 16.0 min. This is one window on one day — treat it as an order-of-magnitude result, not a league table.
Is RSS fast enough to trade on?
On these numbers, no. The median headline took 2.7 min to become visible after its own stated publish time, and the slowest decile took 13.2 min. By then the move is priced. Public RSS is fine for context and terrible for reaction.
Why do day-old articles show up as new in an RSS feed?
Because the feed slot and the publish time are different things. A feed can re-emit an older article — a rewrite, a promotion, a rotation of evergreen content — and it arrives looking exactly like breaking news. In this window 10 of 46 newly-appearing items (21.7%) carried a stated publish time more than six hours old. Nothing in the feed marks them.
Why watch more than one or two feeds?
Because they barely overlap. Of the 45 distinct stories that appeared while we were watching, 100% were carried by exactly one of the 13 feeds. Whichever single feed you watch, most of the tape is happening somewhere you are not looking.