Methodology
How we rate AI prediction tools, tipsters and data providers: two scores, shown side by side and never combined, from public, dated sources.
In short
- Every entry gets two scores, shown side by side and never combined: Performance first, then the Transparency score.
- Performance (0–100, 50 = break-even) comes only from a published level-stakes record, pulled towards 50 when there is little evidence.
- Transparency score (0–100) measures how openly a tool or tipster publishes a record you can verify, from six published criteria. It does not measure profit.
- Data & API providers get a provider Transparency score and an Access & value score and are ranked within each sport; they have no Performance.
- Every figure is sourced and dated, re-checked on a fixed schedule, and every change is on the changelog. Scores are never for sale.
What we rate
We rate AI prediction and tipping tools: products that claim AI- or model-driven selections or predictions. All tool counts, rankings and the scoreboard cover these 122 AI tools only. Data, API and form-data providers (for example The Racing API or Timeform) sell the raw material such tools use but don't publish selections of their own, so they get no Performance score and never appear in the AI tool ratings. They are listed separately on Data & API providers with two scores of their own and a rank within each sport (see Data & API providers below). Human tipsters are rated in their own Tipsters section.
Performance score
Shown next to the Transparency score, never combined with it. It uses only a published level-stakes record (profit/loss or ROI, number of settled bets, period, price basis), with the source linked and the date we checked it. We never estimate missing figures. The same rules apply to AI tools and to human tipsters.
In short: we turn level-stakes ROI into a 0–100 score where 50 is break-even and 100 is +40% ROI, then pull that score towards 50 when there is little evidence. The pull is set by bets ÷ (bets + 500), and self-reported records count as half their bets.
- Raw score
raw = 50 + 1.25 × ROI%capped 0–100: −40% ROI or worse = 0, break-even = 50, +40% or better = 100 - Bets that count
n = settled betsself-reported records count as half (n = bets ÷ 2); independently proofed records and records we recomputed ourselves from the raw picks count in full - Weight
w = n ÷ (n + 500) - Score
score = 50 + (raw − 50) × wrounded to a whole number
Worked examples
Published record: 12,955 settled bets, level-stakes ROI +1.78% (28 Oct 2025 – 28 Oct 2026 (365 days)).
- Raw = 50 + 1.25 × 1.78 = 52.23
- Bets that count = 12,955 = 12,955
- Weight = 12,955 ÷ (12,955 + 500) = 0.963
- Score = 50 + (52.23 − 50) × 0.963 = 52.14 → 52
Published record: 180 settled bets, level-stakes ROI +36.13% (Past 6 months to 30 Oct 2026 (betHQ)).
- Raw = 50 + 1.25 × 36.13 = 95.16
- Bets that count = 180 = 180
- Weight = 180 ÷ (180 + 500) = 0.265
- Score = 50 + (95.16 − 50) × 0.265 = 61.95 → 62 (Provisional: under 200 bets)
How much of the raw score is kept, by sample size
| Settled bets | Weight: independently proofed or recomputed by us | Weight: self-reported |
|---|---|---|
| 100 | 0.17 | 0.09 |
| 200 | 0.29 | 0.17 |
| 500 | 0.50 | 0.33 |
| 1,000 | 0.67 | 0.50 |
| 5,000 | 0.91 | 0.83 |
| 20,000 | 0.98 | 0.95 |
Tiers: Scored, Provisional, Insufficient data, Not assessed
- Scored: 200+ settled bets at level stakes with P/L or ROI published.
- Provisional: 50–199 settled bets. The score is shown greyed out with a small-sample warning and is always sorted below full scores.
- Insufficient data: fewer than 50 bets, no bet count, no level-stakes P/L or ROI, or stakes that vary per pick.
- Not assessed: we haven't checked the record yet.
Badges and flags
- Independently proofed: the record is kept by a third party (for example Tipstrr, the Racing Post Press Challenge or naps table, betHQ).
- Recomputed by us: we rebuilt the figures from the tool's or tipster's own published bet-by-bet data.
- Self-reported: totals published by the tool or tipster itself, or proofed by a company that also sells the service.
- Advised odds, may be hard to match: the record is settled at the price advised or posted at the time of the tip, not at SP or Betfair SP. These prices often shorten quickly or are limited to small stakes.
Sample size and ROI range (tipster pages)
Each tipster page shows the number of settled bets and, where our own data allow it, an approximate 95% range for ROI. It answers: given this many bets, how far could the true return plausibly be from the published figure by luck alone?
- Average winning price
o = (1 + ROI) ÷ strike rate - Standard error of ROI
SE = o × √(p × (1 − p) ÷ n)p = strike rate, n = settled bets - Range
ROI ± 1.96 × SE
Every input comes from the record itself (bets, ROI, winners or strike rate) or from our dated re-check of the same record. Each winner is treated as paid at the average winning price; real prices vary, which only adds spread, so the true range is at least this wide. We don't show a range when no strike rate is published, when stakes aren't level, when the record has no Performance score, or for each-way, place or lay records, where the simple win/lose model doesn't fit. We never estimate a missing input. The range is a guide to uncertainty, not a forecast: a range entirely above 0% doesn't mean future profit. Tipster records are re-checked every 7 days and the page shows the next due date.
Price bases differ (SP, Betfair SP, exchange less commission, bookmaker or advised odds), so scores are not strictly comparable. Subscription fees are not deducted. Past results don't predict future returns.
Tipsters
Human tipsters are rated on the same two scores. Transparency uses the same six criteria and weights, reworded for people. Performance uses exactly the same formula, tiers, badges and flags as for AI tools.
Where tipster records come from
- Where the records come from: Tipstrr service pages (£10 level stakes at advised odds; each-way counts as two stakes), the Racing Post Press Challenge (£1 level stakes to SP; ROI = % of turnover returned − 100) and naps table, betHQ newspaper tables, At The Races monthly reports (recomputed by us from each bet), and Smart Betting Club pages.
- Smart Betting Club services are proofed by SBC, which also sells them, so we treat them as self-reported.
- Facebook pages and groups can't be read without logging in, so those entries rely on the tipster's own sites and listings.
Transparency score
The Transparency score (v0.1, previously called Checkability) measures how openly a tool or tipster publishes a record you can verify. It does not measure profit. Each criterion is either verified (full points) or not yet verified (0 points). Listed yourself? How to raise your transparency score turns these criteria into a checklist, with what counts as evidence for each. The AI tipping transparency report 2026 shows how many of the AI tools and tipsters we rate meet each criterion.
| Criterion | Points |
|---|---|
| Picks published before the off, timestamped | 25 |
| Full public ledger including losers | 25 |
| Price basis stated (SP, BSP, exchange, CLV) | 15 |
| 500+ settled selections | 15 |
| Methodology documented | 10 |
| Results downloadable (CSV/API) | 10 |
| Total | 100 |
Bands
60–100 High transparency30–59 Medium transparency0–29 Low transparency
First seen dates (information only)
Every AI tool, tipster and data provider page shows a First seen date: the earliest dated public evidence that the service existed, with what kind of evidence it is and a link to it. We use, in order of preference, the service's own stated founding or launch date (or the start of its changelog), its Companies House incorporation date (UK companies), the join date shown on its own tipping-platform or X profile, the earliest Internet Archive snapshot of its site, and its domain registration date (RDAP). Where these disagree we show the earliest. If we found nothing, the page says Not found yet. First seen is information only: it is never scored and never changes Performance, Transparency score, Access & value or any rank. Back-tests and historical data from before a launch are fine; only when a live record says it started before the earliest trace do we add a neutral note inviting evidence.
Closed companies and dead sources
We never delete an entry or a source. If a company, tipster or data provider closes or vanishes, its entry is marked Closed with the date we detected it and the evidence, its scores are frozen at our last check (with that date shown), and it leaves every ranking and list. Its own page stays online and it is listed on Gone but on record, so you can see who comes and goes; it follows the same indexing rule as any other entry. A source that stops working (the page itself answers “not found”) is shown struck through and labelled “No longer available”, with the date we last checked it and a link to an archived copy on the Wayback Machine where one exists. Pages that block or time out are never marked dead. A source we stop using is kept as a former source on the entry page.
Which entry pages search engines index
An AI tool or tipster entry with no Performance (Insufficient data or Not assessed) and a Transparency score below 30 (the Low band), or a data provider with a provider Transparency score below 30, has too little verified, sourced detail to be useful as a search result yet. Those pages stay on the site, in every list and fully linked, but ask search engines not to index them (noindex, follow) and are left out of the sitemap. The rule is applied on every build, so a page becomes indexable as soon as the entry has a scoreable record or reaches the threshold. It never changes a score or a rank.
Marketing claims
Claims a tool makes about itself are shown in a labelled box as stated, not verified, until the tool sends evidence through the Is this you? box on its page (it goes to [email protected]). Verified claims are then shown as verified.
Data & API providers
Data and API providers sell the raw material (fixtures, odds, stats, form, racecards) that tools and punters use. They publish no selections, so they get no Performance score. Instead they get two scores of their own, shown side by side and never combined. Every point must trace to the provider's own public page, with the date we checked it. Anything not published, or that we couldn't find, scores 0 and is shown as Not published.
1. Transparency score (providers)
How openly the provider publishes what you get and what it costs. Pricing carries the most weight.
| Criterion | Points |
|---|---|
| Pricing published (a paid price with currency, or a stated free-only product) | 30 |
| Public documentation (readable without signing up) | 20 |
| Coverage list published (sports / competitions) | 15 |
| Plan limits or allowances published (requests, leagues, features) | 10 |
| Public status page | 10 |
| Company identity stated (legal entity or address) | 10 |
| Delivery method stated (API, feed, files, app) | 5 |
| Total | 100 |
2. Access & value score
What you can get, for how little, and in how many ways, measured only from published facts.
| Criterion | Points |
|---|---|
| Free access: ongoing free tier/key/data 20, time-limited trial 10 | 20 |
| Entry price (cheapest published paid plan, GBP): free-only or ≤£10 20 · ≤£40 15 · ≤£100 10 · ≤£400 5 | 20 |
| Data types: 1–2 5 · 3–4 10 · 5+ 15 | 15 |
| Delivery options: REST/HTTP API 5 + push (WebSocket/SSE/webhook) 5 + bulk files 5 + MCP server 5 | 20 |
| Historical data or archive published | 10 |
| Total | 85, scaled to 100 |
The criteria add up to 85 points; we scale the total to 100 and round to the nearest whole number. Sports covered is shown on each provider page as information only and earns no points (since 21 Oct 2026): a single-sport specialist isn't penalised for its focus, and depth within a sport is already captured by Data types and Historical data.
Entry price uses the cheapest published paid plan: monthly plans at their monthly price, annual-only plans at the monthly equivalent, one-off purchases at face value. Other currencies are converted to GBP at the ECB reference rates for 16 Oct 2026 (USD 1.1253, GBP 0.84385, AUD 1.6171 per EUR). A product that is free with no paid plan gets full entry-price points. If no paid price is published, entry price scores 0. Status-page uptime is not scored yet; we plan to add it once we have 90 days of readings.
Bands
60–100 High30–59 Medium0–29 Low
Ranking
Providers are ranked within each sport or category (for example Football, Horse racing, Odds feeds, Betting exchanges, Historical datasets). A provider can appear in several. Order: Transparency score (high to low), then Access & value score (high to low), then name. Two providers with the same two scores share a rank, shown as “=3”. The scores are never added together.
Coverage regions
The location filter shows where a provider's data covers, not where the company is based: USA if it lists NFL, NBA, MLB, NHL, MLS or NCAA; UK & Ireland if it lists UK/IE racing, the Premier League, EFL, Scottish football, or UK/IE darts, snooker or cricket competitions; Australia for AFL, NRL, Australian racing or competitions; International for anything else. A provider can be in several regions. Each region links to the page where we saw the competition named.
What the scores don't say
- They don't measure data accuracy, latency or uptime. We haven't tested those.
- They are not an endorsement or a recommendation to buy.
- Sponsored listings sit outside the rankings and are labelled “Sponsored”, in line with CAP Code and ASA rules on identifying marketing communications. Sponsorship, advertising and affiliate links never change a score, a rank or where a provider appears in the ranked list.
- Providers can send corrections with a public link; we re-check and log every change. What counts as evidence: how to raise your transparency score.
Rules
- Only the tool's own public pages and data; every figure is sourced and dated.
- We never publish our own selections.
- Scores are never for sale; paid placements are labelled “Sponsored”.
- Vendors can request corrections; changes are logged.
Corrections
If a score or figure is wrong, email [email protected] with the page and a link to the public evidence, or use the Is this you? box on the entry page, which prefills that email for you. We review every request against the published criteria and log changes. What counts as evidence: the transparency checklist. See our independence and corrections policy. Our ratings are also available as an open data file (CSV, CC BY 4.0).
How often we re-check
- Daily, automatically: records with a machine-readable public results page (for example Tipstrr public profiles) are re-read every day, and the figures and check date update only when the page was actually fetched and read.
- Weekly, by hand: claims, pricing and ownership are reviewed every week, along with records we can't read automatically.
- Records no longer fully public: if a record we verified stops being fully public (for example a profile that now shows only a recent window), we keep the figures we last confirmed publicly, show that date and a neutral note inviting the owner to republish the full record, and keep re-checking; we never replace a full record with a shorter window. If the full record is still not public 30 days later, the Transparency criteria that rest on it (such as Full public ledger including losers) lapse and their points are removed; they count again once the full record is public again.
- Anything not checked in the last 7 days is flagged Due for re-check. An entry we haven't verified yet says Not yet checked.
Every entry shows Numbers last checked (when we last verified its figures and claims against the linked sources) and Entry updated (when its data last changed). Site data last updated: 5 Nov 2026.
Every change to the ratings (new entries, method changes, reclassifications, corrections and automatic re-checks that changed a figure) is listed with its date on the changelog.