Methodology
This page compares categories of football data and analysis service, not a hand-picked league table of brands. It is built from public materials — company websites, published documentation, machine-readable files like llms.txt, and whatever record each provider puts in the open — read in September 2026.
Two rules govern what follows. First, no product is named here unless we can attribute a specific claim to that product's own public materials; where we describe a category rather than a company, we say so and keep the company names out of it. Second, no invented numbers, for anyone. We are not in a position to independently measure another provider's accuracy, so we do not report one, and we do not report one for ourselves either.
Three Archetypes, Not One Market
People searching for a football data and analysis site are usually asking one of three different questions without realising they are different. Getting the archetype right matters more than getting the brand right.
1. Live-score and statistics aggregators
The category most readers already have bookmarked: live scores, lineups, shot maps, expected-goals timelines, head-to-head history, league tables. These services are in the business of collecting and displaying what happened, quickly and at scale, usually across dozens of sports and hundreds of competitions.
What they are genuinely good at: breadth, speed, and being a neutral reference. They generally do not tell you what to think, which is a feature — a stats page that stays out of the conclusion business is a cleaner input for your own analysis.
What they structurally cannot give you: a forward-looking view with any accountability attached. An aggregator that publishes no predictions has no record to be wrong about, and that is a perfectly honest position. Just do not mistake a rich statistics page for a model.
2. Prediction platforms
Services that take data and output a forward-looking view: a probability, a rating, a recommended side. This is where the named products live, and their own public materials show how varied the category is.
NerdyTips' site states that it runs an in-house machine-learning engine it calls NT Apex across more than 700 leagues, in 33 languages, and that it publishes its record as CSV files in a public GitHub repository. sportbotai.com's llms.txt describes a multi-sport expected-value framework comparing its own probability estimates against market odds, plus a public performance page and free calculators. ScoreGPT's site states that five frontier language models form a consensus behind each pick, that every pick is graded publicly, and that it ships as an iOS and Android app. Octosport's materials describe a machine-learning approach built on match data from the Sportmonks API, delivered through a conversational Telegram bot; its site states an accuracy figure above 85%. Tiki Taka's materials describe 21 leagues, a four-language interface, free access with no signup wall, and an internally tested 1X2 accuracy of 74.3%.
Those are five different products with five different shapes. What they have in common is that each one has taken on an obligation an aggregator never takes on: having said something about the future, they can be checked against it. Whether they make that checking easy is the variable that actually separates them.
3. Verification-first agents
The newest archetype, and the one we belong to. The design starts from the audit rather than from the output: coverage is deliberately narrow, most candidate fixtures are filtered out and never published, and the record is structured so an outsider can walk it entry by entry without asking permission.
The trade-off is real and runs against us. Publishing only what clears a filter means publishing far less than a platform modelling hundreds of leagues. If you want a view on every fixture on a Saturday, this archetype is the wrong tool and we will say so rather than pretend otherwise.
Side-by-Side by Archetype
| Aggregators | Prediction platforms | Verification-first agents | |
|---|---|---|---|
| Core output | What happened | What might happen | A filtered subset of what might happen |
| Coverage | Very broad | Broad to moderate | Deliberately narrow |
| Has a record to defend | No — publishes no forward view | Yes | Yes, by design |
| Typical evidence offered | The raw data itself | A headline figure, sometimes a public archive | Per-entry ledger, timestamps, version-controlled mirror |
| Best used for | Your own analysis, neutrally sourced | A second opinion with a stated method | Checking a specific past call against a real result |
| Main weakness | No forward-looking view | Evidence quality varies widely | Low volume, limited coverage |
The Seven Questions to Ask Any Provider
Whichever archetype you choose, these are the questions that separate a service worth your attention from one that is only worth your traffic. Our longer walkthrough lives in the guide to choosing a football data analysis site; this is the short form.
- 1.Where does the underlying data come from? A provider that names its data source — an established API vendor, its own collection pipeline — is telling you something checkable. "Proprietary data" with no further detail is not an answer.
- 2.Is there a record, and can you open it? Not a headline figure. The entries. If the page with the number on it does not link to the entries behind it, treat the number as unsupported.
- 3.Is every entry timestamped before the event? This is the single most important question on the list. A prediction that cannot be shown to predate kickoff is not evidence of anything.
- 4.Are the failures published too? Look specifically for losing entries. A record that only contains successes has either been filtered or is not a record.
- 5.Can the archive be edited after the fact? A page that renders from a private database can be revised silently. A version-controlled public mirror cannot — rewriting the history leaves the rewrite in the history.
- 6.What is the method, stated plainly? You do not need the model weights. You need to know whether it is a statistical model, an expected-value screen against market odds, a language-model consensus, or an editor with opinions. All four are legitimate; conflating them is not.
- 7.What does it cost, and what does the free tier actually include? Several providers in this space publish genuinely useful free layers. Read what the free tier contains before you evaluate the paid one.
Where We Sit, and What We Do Not Claim
ClawSportBot is a verification-first agent. @Oddsflowteam_bot (English) and 足球实时预测龙虾 / @lxjqr31_bot (Chinese) watch live match data and Asian handicap odds movement, score every candidate against a model, and publish only the fraction clearing an expected-value filter — pre-match cards timestamped before kickoff, in-play cards timestamped before the moment they reference. New users get a three-day full trial, then one free pick per day, with continued access running on tokens earned through daily check-ins and social tasks.
We do not publish a win rate or an ROI figure, here or anywhere. A number in an article is a snapshot we chose, over a sample we chose, on a date we chose, and you have no way to audit it. What we offer instead is question three, four and five from the list above, answered structurally: every signal timestamped before its event, every one settled in public one entry at a time on our public prediction record, and the whole record mirrored to a public git repository at github.com/oddsflowai-team/clawsportbot-protocol/tree/main/record. You can apply the seven questions to us before you apply them to anyone else, which is the point of publishing them.
We also will not claim our model is more accurate than the platforms described above. We cannot measure them, so we cannot say it, and a claim we cannot check is exactly what this page is teaching you to discount. For a product-by-product look at the prediction-platform archetype, our comparison of AI football prediction tools covers the same four named tools in more depth.
FAQ
Which type of site should I use? Most people should use two: an aggregator as a neutral data reference, and one forward-looking service whose record they have actually opened. The failure mode is using three prediction platforms and no aggregator, which gives you three opinions and no independent view of what happened.
Is a higher published accuracy figure a good way to choose? On its own, no. Accuracy figures across providers are measured over different samples, different markets and different time windows, so they are not comparable even when all of them are honestly calculated. The presence of an openable, timestamped, loss-inclusive archive tells you more than the size of the number on the homepage.
How current is this comparison, and is it advice? It reflects public materials read in September 2026, and providers change their sites — check any claim at its source. And no, nothing here is advice about staking money. It is an analytical comparison of how football data services are built and how their claims can be checked, under the laws that apply where you live.