Guides · For brands and retailers
Fake clicks, bots, made-up sales: how do I check the figures when creators are paid on results?
If you pay on clicks, views or sales, check every metric against a second source: clicks against sessions in Google Analytics, views against reach and watch time, reported orders against your shop data. The Media Rating Council’s standard separates easily filtered invalid traffic from the hard-to-detect kind, and a gap on its own proves no fraud. Write down your limits in advance, in your own programme or when you set KPIs before launch with wyrld Boost.
By the wyrld team · Last updated: · 9 min read
With wyrld Boost, many creators feature your products at the same time on Instagram, TikTok or YouTube. Your budget is fixed up front, and the creators are paid from it by results.
Key points
- Google Analytics excludes known bots automatically. Google says you can neither turn this off nor see how much traffic it affected.
- The Media Rating Council separates general from sophisticated invalid traffic: lists catch the first, the second needs analysis and includes click farms.
- Assume 1,000 clicks, 620 sessions and 75% consent: the banner accounts for about 207 visits, leaving just 173 for failed loads and bots.
- Zero errors in 20 random orders only means the error rate is below about 14% with 95% confidence. A 5% limit takes 59 clean orders.
On this page
- How easily views, clicks and sales can be faked
- What the Media Rating Council standard distinguishes
- Ten plausibility checks for creator figures
- Worked example: 1,000 clicks, 620 sessions, 4 orders
- Sample check: 20 reported orders against your shop data
- What should be in writing before the first post
- Where the figures come from with wyrld Boost
How easily views, clicks and sales can be faked
| Metric | Exposure | Typical manipulation | Cross-check in your data | Limit to set in advance |
|---|---|---|---|---|
| Views | High | Bought views | Watch time, reach, earlier posts | Minimum duration per view |
| Clicks | High | Bots, link previews, click farms | Sessions, time of day, country, device | What counts as a click |
| Watch time | Medium | Endless loops, bought views | Analytics export per post | Who supplies the export |
| Sales via link | Medium | Self-purchases, later cancellations | Order number, address, return status | Cut-off after the return period |
| Sales via code | Medium | Code posted on voucher sites, self-purchases | Order time against post time, a web search for the code | Who may use the code |
The further a metric is from the money, the cheaper it is to fake. Bought views cost little; a real order costs the value of the goods, at least to begin with. The ratings in the table are our own assessment based on how the measurement works, not the result of a study.
The platforms ban such methods: YouTube prohibits anything that artificially inflates views, likes or other metrics, and TikTok prohibits trading in services that artificially boost engagement. But a ban is no substitute for a cross-check, because what you pay for is whatever ends up in the report.
What the Media Rating Council standard distinguishes
The US standard from the Media Rating Council (MRC) divides invalid traffic into the kind that is easy to filter and the kind that is hard to detect. General invalid traffic (GIVT) can be found with lists and routine checks: known data centres, bots and crawlers, for example. Sophisticated invalid traffic (SIVT) needs advanced analytics, several points of corroboration or manual work. The MRC includes hijacked devices, click farms run by people and falsified measurement events, up to and including conversions.
The June 2020 paper was written for ad measurement, not for creator links, but the distinction still helps. Google Analytics excludes known bots and crawlers automatically, and Google says you cannot see how much was affected. Anything not on a list can remain in your sessions. For the second group, then, your own cross-checks are all you have.
Ten plausibility checks for creator figures
| Comparison | Warning sign | Check | Data source |
|---|---|---|---|
| Clicks to sessions | Far fewer sessions per click than for other creators | Set the ratio per creator side by side | Link tool, analytics |
| Sessions to baskets | Many sessions, hardly any baskets | Compare the add_to_cart event per creator | Analytics, shop |
| Visits by time of day | Spikes with no post, an even flow overnight | Plot clicks per hour against the post time | Link tool export |
| Visits by country | Lots of visits from countries with no audience or no shipping | Country report against the creator’s audience | Analytics, creator statistics |
| Visits by device | Almost all desktop, though the post runs in a mobile app | Check the device report per creator | Analytics |
| Views to followers | A jump far above earlier posts | Compare the same person’s last ten posts | Creator statistics |
| Views to watch time | Many views, very short playback | Ask for the average view duration | Creator statistics |
| Self-purchases | Same delivery address as the creator | Match in-house against the creator’s address, clear data protection first | Shop system |
| Codes without clicks | Code sales with no post, code on voucher sites | Order time against post time, search the web for the code | Shop system, search engine |
| Cancellations and returns | Rate far above the shop average | Status per order after the return period | Shop system |
Creator figures are plausible when they match a second source and the other creators in the same campaign. We have not found a generally accepted percentage threshold for creator links. So use the campaign median as your benchmark and write down beforehand how far from it a creator has to be before you ask, for example half as many sessions per click as the median.
A warning sign is a reason to ask, not proof. A video that travels a long way in another country, or a slow page, can look much like automation in the figures. And don’t compare periods across a change in measurement: since 24 August 2026 YouTube has counted the start of playback as a view, so figures from before and after that date do not measure the same thing.
Worked example: 1,000 clicks, 620 sessions, 4 orders
| Item | Count | How to work it out |
|---|---|---|
| Clicks in the link tool | 1,000 | Export with timestamps |
| minus bots and automatic previews | 110 | Bursts seconds apart, known identifiers |
| minus failed page loads | 63 | What is left of the gap |
| minus visits without analytics consent | 207 | 620 divided by 0.75, minus 620 |
| Sessions in Google Analytics | 620 | Report by source or UTM |
| Orders in the shop | 4 | Shop system, counted even without consent |
A gap of 380 clicks is not proof of fraud on its own, because it splits into several causes and only one of them is automation. All figures here are assumptions. If 75% of visitors consent to analytics, 620 sessions represent about 827 real visits. Google says Analytics sets no cookies of its own without consent; with advanced consent mode it models part of the gap, and the calculation changes.
In the example, what remains stands for pages that never finished loading; in real data it can also hide automation nobody caught. Meta says it introduced landing page views in 2017 because of exactly such failed loads, since slow sites and poor connections make many people give up while they wait. For payment, the definition decides: 1,000 raw clicks, 890 clicks without automation, or 620 sessions. A creator has little say over consent or page speed, which is why the middle figure is often the fairest basis. It just has to be agreed before launch.
Sample check: 20 reported orders against your shop data
The fastest way to check reported orders is a random sample against your shop system, and it is worth doing before any payout that covers many orders. A clean sample is no acquittal, though: no discrepancies in 20 orders only means the error rate is below about 14% with 95% confidence (1 minus 0.05 to the power of 1/20). For a 5% limit you need 59 error-free orders.
What counts as a self-purchase is your call. Amazon’s Associates programme policies prohibit purchases through your own affiliate links, including by friends, relatives or staff. In your own programme the rule can be different; it just has to be set in advance.
- Export every reported order with order number, date, amount and creator ID.
- Pick 20 orders using random numbers, not the biggest or the most recent. If there are fewer than 20, check them all.
- Check in the shop system that the order number exists and that the amount and date match.
- Check the status: paid, cancelled or returned.
- Check where it came from: was the order placed after the post, and is there a link or code attached to it?
- Match the delivery address in-house against the creator’s address. No customer data leaves the company, and because this is personal data, agree the check with your data protection adviser beforehand.
- Record discrepancies for each check. If you find one, widen the sample or check every order for that creator.
What should be in writing before the first post
Checking rules only stand up if they exist before the first post; afterwards, any limit looks like an excuse. Six definitions are enough. Our order: these six first, then launch, then the sample check before every payout. What we would not do is suspect a creator because of a single odd figure.
- Click: raw click, click without automation, or session in the shop.
- View: start of playback or a minimum duration. Since 24 August 2026 YouTube has counted from the start of playback; TikTok’s ad metrics report 2 and 6 seconds separately.
- Data source: who supplies which export, with timestamps and in what format.
- Sale: an order that has been neither cancelled nor returned by the end of the return period.
- Self-purchase: who is included, for example the creator, their household or their team.
- Checking threshold: how far from the median a creator has to be before you ask.
Where the figures come from with wyrld Boost
With wyrld Boost you set the KPIs before launch, which is the moment to raise definitions such as click, sale and self-purchase. wyrld tracks clicks through the links, the shops report sales, and creators upload watch time and reach from their own statistics. If people buy in your own shop, you hold the orders anyway. How far reporting goes beyond results per creator depends on the package, so check beforehand what your cross-checks need. It is the wrong route if you need a promise on reach or revenue.
wyrld Boost at a glance
- What it is
- The campaign service from wyrld, a social commerce platform based in Hamburg, Germany. Many creators feature your products at the same time, so you do not have to book each one. People buy in your own shop, on Amazon or through price comparison: you decide.
- Price
- You set the budget. Campaigns usually start from €10,000 plus VAT. Creators are paid from the budget according to the KPIs they reach, such as purchases, clicks or watch time. The package price also covers selection, payouts, reporting and a short consultation.
- Creators and videos
- wyrld selects suitable creators, and each creator decides whether to take part. Creators label campaign videos as advertising. The videos run on the creators’ channels. The right to reuse them yourself is not part of the package.
- How it runs
- Depending on the campaign, you approve the videos before they go live, or the creators post from your brief. We set out the minimum scope of the campaign in your offer. We do not promise reach or revenue. After the campaign you can see which creator delivered which results.
- Who it is for
- wyrld Boost is open to brands of every size and in all categories, including brands from abroad that want to sell in Germany, Austria or Switzerland, provided the product and its advertising are lawful. Agencies book as normal customers.
- Brand and shop subscription
- Separately, this shows you every day which creators feature your products and, where the shop reports purchases, who is selling. Price on request.
Frequently asked questions.
How can I spot fake clicks from influencers?
Compare the clicks in the link tool with the sessions in your shop, creator by creator, looking at time of day, country, device and baskets. Spikes with no post and sessions with no action at all stand out. A single value proves nothing; only several deviations together give you reason to ask.
Does Google Analytics filter out bots automatically?
Yes. Google Analytics automatically excludes known bots and crawlers, based on Google’s own research and the Interactive Advertising Bureau’s bot list. You can’t switch this off, and Google doesn’t show how much was excluded. Unknown bots and fake clicks generated by people can stay in the data.
Why does my shop show fewer sessions than the link tool shows clicks?
Usually several causes combine: bots and automatic previews, pages that never finish loading, and people who decline analytics. Only breaking the gap down shows how much of it is automation. A gap on its own is no proof of fraud.
How many orders should I check in a sample?
Twenty orders drawn at random are a practical start. If all of them are clean, the error rate is below about 14% with 95% confidence. For a 5% limit you need 59 clean orders. If there are only a handful of orders in total, check every one.
Is buying through your own link fraud?
That depends first on your own rule. Amazon prohibits such purchases for its affiliates; in your own programme you decide and put it in writing beforehand. No metric can tell you if something is fraud in the legal sense, so if in doubt, take legal advice.
Last updated:
Sources
This page explains checks in general terms and is not legal or data protection advice. Accessed on :
- Media Rating Council: Invalid Traffic Detection and Filtration Standards Addendum (June 2020)
- Google Analytics Help: Known bot-traffic exclusion
- Google Analytics Help: Consent mode
- Meta: Metrics Release (June 2017)
- YouTube Help: How engagement metrics are counted
- YouTube Help: Fake engagement policy
- TikTok: Community Guidelines, Integrity and Authenticity
- TikTok Ads Help: Video play metrics
- Amazon PartnerNet: Associates programme policies