OddAlerts AI football predictions: what the free model gives you and what Pro unlocks

Our rating

What OddAlerts is and who runs it

OddAlerts belongs to Made By Joe LTD, a company registered in the United Kingdom under number 08349127. The site started in 2020 as an alert tool. You set a rule such as “tell me when Manchester City goes above 2.00”, and the site sent a message. Since then it has grown into a big platform with more than 50 modules: stats tables, expected goals, a bet tracker, odds movements from Pinnacle (a bookmaker whose prices are seen as the most accurate in the market), player props, a Polymarket page and an API for developers.

Contact is by email only. The address is hidden behind a click-to-reveal link on the Terms page, and there is no phone number or postal address anywhere on the site. Outside email, the site talks to users on X (@oddalerts), a Telegram channel, a YouTube channel and a Discord server for paying members.

The AI predictions page is one of those modules. This review looks mainly at that page and at the Alpha results page, because that is what matters for this category: does the model work, and what does it cost.

The Trustpilot profile was claimed in March 2025 and has one review. That is too little to judge anything. The results page on the site itself is the only place where you can check the service by numbers.

What the free AI model shows on every match

Open the “AI Football Predictions” tab and you get the full list for the day. On 9 September it had 246 matches, from the Czech under-19 league to the Ukrainian Cup. For each match the model gives three numbers in percent: home win, draw, away win. The biggest one is shown in green.

Click a match and a card opens. On a free account you see:

• Match result: home, draw, away in percent.
• Both teams to score, yes or no. This market asks a simple question: will each team score at least one goal?
• Over 2.5 and under 2.5 goals. “Over 2.5” means three goals or more in the match. “Under 2.5” means two or fewer.
• The most likely final score.
• One “edge” line. It shows the model's pick, the bookmaker price and the gap between them. For Opava U19 against Ceske Budejovice U19 it said: “Opava U19 to win @ 3.00, model 33.8% vs 33.3% implied, +0.5%”.

What “implied” means. Every odds number hides a probability. Odds of 3.00 mean the bookmaker thinks the event has about a 1 in 3 chance, so 33.3%. That number is the “implied probability”. When the model's number is higher than the implied one, the site calls the gap an “edge”.

The odds in the edge line are the bookmaker prices OddAlerts collects, mainly from Bet365 and Pinnacle. Your own bookmaker may show a different price. An edge of +0.5% at 3.00 disappears if your book offers 2.90. Before you act on an edge, compare the price on our odds comparison page.

Everything below the edge line is blurred: other markets such as handicaps, team totals and the exact final score. The “Edge” sort button and the “minimum probability” slider are locked too. So on a free account you can read the model, but you cannot sort the whole day by the biggest gap between the model and the market.

Sign-up takes an email, a Google account or a Telegram account. No card is asked for.

Bayesian model, not a neural network

OddAlerts calls its engine a “probability model” and describes it as “Bayesian predictions for every match”. In plain words: the model starts from past results and updates its numbers every time new data comes in. It is a statistical method, older than neural networks, and it has been used for football for a long time.

This matters when you compare services. NerdyTips says it uses a neural network. BetMines uses a mix of statistical models. OddAlerts uses a Bayesian model and puts a text layer on top of it: the Copilot reports. The site does not say what writes them, but the format and the wording look like a language model working from the same numbers. So the word “AI” on the tab covers two different things. The percentages come from statistics. The text is written on top of them.

The percentages come first, the text is written from them. When the two disagree, the percentage is the original. But the percentage itself can be wrong, and the test below shows where.

We could not find a page that shows how the probability model did over time. There is no “yesterday” page with hits and misses, and no chart that compares the model's percentages with real results. That is the biggest gap in the product, and it is why the transparency score above is not higher.

The 9 September “biggest edges”: where the model's numbers come from

At the top of the predictions page there is a row of cards called “biggest model edges”. These are the matches where the model disagrees most with the bookmaker. We wrote down the three largest ones on the morning of 9 September.

Three matches is not a test of accuracy. It is a look at where the model's numbers come from, and you do not need the final scores to see the problem. Two of the three were settled by the time we published. Napoli U19 lost 1-2 at home to Arsenal U19, so the 83.8% pick failed. Opava U19 won 4-2, so the +0.5% pick won at 3.00. The South Hobart match was not listed on the results services we checked, and we will add it when it is.

An edge of 84% against 9.1% is not a normal edge. A bookmaker that prices a women's team in Tasmania at 11.00 knows something the model does not, or the model has almost no data for that league. The same is true for Napoli U19 at 83.8%. When the model is that far from the market in a small league, treat the number as a data problem and not as a chance to make money.

A bookmaker gives odds of 11.00. The model says 84%. One of them is wrong, and in small leagues it is usually the model.

The Opava U19 line shows the other side. There the model agrees with the market almost exactly, and the card says so: +0.5%. That kind of honesty is useful. The site does not push every match as a bet.

The Copilot reports do not always agree with the numbers either. On 9 September the Fortaleza report said the model “backs a home win”, while the edge line for the same match showed minus 3.6%. When the text and the number differ, follow the number.

Copilot Classic: 3 free reports a day

Each match has an “AI Match Report” button. On a free account it first opens a sales screen for Copilot 2.0. Scroll down and there is a small link: “Run Classic with a free analysis instead”. That gives you 3 reports a day.

We ran it on Gwangju against Jeju United in the Korean K League. The report always has the same structure:

• Tags for each team, such as “Leaky Defence”, “Fades Late”, “Strong Away”.
• A short preview with head-to-head numbers, that is, past matches between these two teams.
• One paragraph per team: goals per game, home or away form, how often they finish a match without conceding.
• A verdict.
• The most likely outcomes.
• Value Bets. For this match it said: “No genuine value – all model probabilities fall short of implied odds by more than 3%”.
• A bet builder with an approximate price. A bet builder puts two or more conditions from one match into a single bet. Here: over 1.5 goals plus a first-half goal at about 1.75.
• An “Avoid” line: under 1.5 goals, model 28.5% against 79% implied.
• Key stats.

The first half of the report is a match preview. The numbers in it are the same ones the model uses: goals per game, form at home and away, how often each team concedes two or more. The second half is where the report becomes useful for a bet.

The “no genuine value” line is the part we liked most. Many AI tip sites give you a bet for every match. This one tells you there is nothing to bet on when the numbers say so.

“No genuine value” is the most useful sentence a prediction tool can write. It means: do not bet on this match.

The follow-up chat box at the bottom is locked on the free plan. Pro opens it, so you can ask the report a question about one market or one team. Copilot 2.0, which also comes with Pro, adds team news, the referee and a “how to play it” section with prices. Pro gives 30 reports a day. Advanced has no limit.

Five minutes with OddAlerts: a routine for a beginner

This is the routine we would give to someone who has never used a prediction tool. It uses only the free plan.

1. Open the “AI Football Predictions” tab and click the day you want to bet on.
2. Type your team into the “Team” box on the left. The list shrinks to that match.
3. Read the three percentages. Then open the odds at your own bookmaker and turn them into a percentage: divide 100 by the odds. Odds of 2.50 are 40%. Our odds converter does this for you.
4. Compare. If the model's number is clearly higher than your bookmaker's number, the site sees value. If it is lower, the site sees no value, whatever the text says.
5. Spend one of your three free Copilot reports on that match and read the “Value Bets” and “Avoid” lines.

Two small things to know. The site is in English only. All kick-off times are shown in London time, so a 19:45 match is 21:45 in Nairobi and 00:15 the next day in Delhi.

What is OddAlerts Alpha and how is it different from the AI predictions?

Alpha is a separate page. It is a daily list of football bets picked by “systems”: sets of rules that were tested on years of real bets placed by OddAlerts users. Each pick is published before kick-off with the price at that moment. After the match the result goes into a public table. Losses stay in the table. Postponed matches are cancelled and the stake is returned.

On 9 September 2026 the page showed:

• Profit: +332.3 units.
• ROI: +3.0%.
• Settled bets: 11,148 since 22 April 2026. “Settled” means the match is over and the bet has been counted as a win or a loss.
• Strike rate: 37.3%.
• Average odds: 3.33.

Units, ROI and strike rate in plain words. A “unit” is one fixed stake, for example $10. “Flat 1-unit stakes” means every bet is the same size. “ROI” is profit divided by the total amount staked: +3.0% means $3 back for every $100 put in. “Strike rate” is the share of bets that won. 37.3% at average odds of 3.33 is a normal pair of numbers: high odds, fewer wins.

By market, the biggest volume is under 2.5 goals: 3,903 bets at +3.1%. Home wins come second with 3,550 bets at +2.0%. Away wins, draws and under 3.5 goals are all between +8% and +9%. Two markets lose money: under 1.5 goals at minus 25.5% on 35 bets, and over 4.5 goals at minus 13.4% on 241 bets.

Alpha is not the AI model. The AI page gives a probability for every match. Alpha picks a small number of bets using rules found in old data. The public record belongs to Alpha only. When another review says “the OddAlerts AI has a public 3% ROI”, that is a mix-up. The AI model has no public record. Alpha does.

The AI page and the Alpha page are two different products. Only Alpha has a public results table.

On a free account you can see the whole results table and the list of upcoming matches, but the pick and the odds are locked. Alpha picks open with the Advanced plan or the Season Ticket.

+159 units in 30 days against +332 units overall

The headline number needs one more look. The whole record since April is +332.3 units. The last 30 days alone gave +159.3 units at +6.0% ROI on 2,670 bets. So almost half of the total profit came in the last month.

There is a reason for this. On 23 July the site replaced the first version of the system with a stricter second version. Since then the current version has made +178.9 units at +4.8% on 3,740 bets. The first version, from April to July, made money but stayed flat for long stretches. You can see it on the chart: the line moves near zero until July, then climbs.

A smaller group called “high conviction”, the picks where the system disagrees most with the bookmaker, was fixed on 14 July. It has 638 settled bets at +6.1% ROI. That is a small sample. At average odds above 3.00, a run of 638 bets can move by tens of units in either direction just by luck.

Four and a half months is a real record. The strong part of it is seven weeks old. Wait before you call +4.8% a stable number.

Our reading: the site deserves credit for publishing every loss, and the record is real. But the good part of the curve is short. If you had subscribed in June, you would have spent three months close to zero. Come back in six months before you treat the current ROI as normal.

How much does OddAlerts Pro cost?

Prices are in British pounds. The dollar amounts are approximate at £1 = $1.36 on 9 September 2026.

Pro opens all goal lines and correct scores on the AI page, the edge sort, the filters, the bet tracker, the full stats tables, live odds alerts and 30 Copilot reports a day. The 24-hour pass is the cheapest way to test it. The pass cannot be refunded.

One inconsistency: the plan window shows 6 months at £89.99, while the checkout page offered 6 months at £99.99. Check the final price before you pay.

Advanced at £69.99: who actually needs the API

Advanced adds the Alpha picks with alerts, an API for developers, five value-bet alert profiles instead of one, faster Pinnacle alerts, Polymarket tracking and unlimited Copilot reports. If you bet from the website and do not write code, Pro plus the free Alpha table is enough. Advanced is for people who want the picks sent to their phone before kick-off, or who build their own tools on the API.

The site also sells a Season Ticket at £179.99 (about $245) for 12 months. It is the Pro plan paid for a year, so it works out at 12 months for the price of 9, and it adds early access to features in testing. According to the Alpha page, Season Ticket holders also get the Alpha picks and alerts, even though the ticket is not the Advanced plan. Sales close at the start of each new season.

Value Bets, In-Play Funnel and the Telegram condition

The Value Bets tab is open on a free account. It works as a small app marked “beta”. You pick a strategy, for example “Home Win Strategy”, and the feed lists matches where the model's chance is higher than the Bet365 opening price, the price the bookmaker gave when the market first opened. Each card shows the live probability, the implied probability, the odds, the value in percent and a “Predictability” label from Poor to High.

What a value bet is. A bet where the real chance of winning is higher than the chance hidden in the odds. If the model says 60% and the odds say 50%, the site calls it value. Whether the model is right is another question.

The In-Play Funnel is the live alerts tool and part of the Pro plan. It scans odds every second across more than 2,500 leagues and sends an alert when a live match matches your rules. It does not run without Telegram. The first screen asks you to link the OddAlerts bot, and there is no email or browser option.

The Suggested Alerts page is also free and shows full stat rows for each suggested bet: first-half goals, cards, corners. It is more than we expected from a free plan.

The bet tracker is Pro only. If you want a free tracker, our bet trackers section has options.

Refunds within 10 days, but not for the passes

The refund policy, updated on 9 September 2026, gives a full refund on any payment made in the last 10 days. You write to support, give your account details and the reason, and the money comes back within 5 working days to the same payment method.

The 24-hour Pro pass and the 48-hour Advanced pass are excluded. They are sold cheaply for testing, so they are final.

You cancel in Settings. Access stays until the end of the paid period, and no further charge is taken. Your filters and alerts are kept in case you come back.

Payment goes through Stripe: card, PayPal, Revolut Pay, Apple Pay and bank debit. The checkout can show prices in Turkish lira as well as pounds.

No app in the stores

There is no OddAlerts app in the App Store or Google Play. The “Install OddAlerts” button in Settings adds the website to your phone screen as a web app. Push alerts for Alpha picks go through that web app. Live alerts go through Telegram.

OddAlerts against NerdyTips and BetMines

Read our NerdyTips review and BetMines review for the full numbers. The short version: OddAlerts gives the widest free view of a model and the only real refund. NerdyTips gives the biggest results file, and that file shows a loss. BetMines gives the clearest day-by-day page, and its own numbers show the same.

11,148 settled bets at +3.0%: what the record proves and what it does not

The record proves that OddAlerts publishes its picks before the match and keeps the losses. That is rare, and it is the reason to trust this site more than most AI tip pages.

It does not prove that the AI predictions page is accurate. The two are different products. The percentages you see for free come from a model with no public scorecard. The +3.0% belongs to a separate feed, and its picks are open only with Advanced, the 48-hour Advanced pass or the Season Ticket.

Use the free plan to check your own bets. Pay only if you will use the alerts or the tracker every week.

The best way to use OddAlerts is the free plan. Check your own bet against the model percentage. Run the three free Copilot reports on the matches you care about. If the model says “no genuine value”, listen. If it says 84% on a team priced at 11.00, do not.

Pay for Pro if you use the stats tables, the tracker or the live alerts every week. Pay for Advanced only if you want Alpha alerts on your phone or the API.