The Meta Dataset Quality API is the developer interface that reports, for each match key, how well the customer information in your Conversions API events identifies a Meta account. It sits underneath the single Event Match Quality score and names the specific problem: an empty IP address, a placeholder email, a click ID missing from most events. Read it before you change an event payload, because it tells you which field to fix.
What the score behind it measures
Event Match Quality is Meta’s number from 0 to 10 for one event, and it combines two things: the quality of the customer information sent with that event, and the share of those events Meta managed to match to an account. It exists for website events sent through the Conversions API with action_source set to website, standard or custom, and it is calculated on the last 48 hours of data. Meta’s own guidance is to aim for 6.0 or higher.
The score matters because of what happens to an event Meta cannot match. Matched events feed attribution and delivery optimisation. Unmatched ones count for basic measurement and teach the delivery system nothing. In lead generation that is the whole game: a Lead nobody can match is a lead the system never learns from.
What the API adds beyond the score
A score of 4 tells you something is wrong. The Dataset Quality API tells you what. It reports Event Match Quality diagnostics per match key, and in September 2026 Meta added six issue types to that report. Read together, they cover the mistakes a lead generation setup makes most often:
| Diagnostic | What it says | Usual cause on a landing page setup |
|---|---|---|
| Empty IP | events arrive with no client_ip_address | the server never copies the visitor’s address into the event |
| Server sending mismatched IP addresses | the server addresses do not match those the Meta Pixel saw | your own server or proxy address is sent instead of the visitor’s |
| Relay IP | the address belongs to a privacy relay, not the customer | visitors browsing through a relay, so the rest of the event has to carry the match |
| IPv4 | an IPv4 address was sent where IPv6 would match better | a landing page domain that answers over IPv4 only |
| FBC coverage | fbc is present on too small a share of events | fbc read only from the cookie, never built from the fbclid in the URL |
| Email placeholder | many events share one placeholder email | a default value filled in when the field was left empty |
| Phone placeholder | many events share one placeholder phone number | the same habit on the phone field |
Two of these are visible only here. A placeholder email produces events that look complete in Events Manager and match nobody, and weak FBC coverage stays invisible until you put the share of events carrying fbc next to the share carrying email.
What good data looks like on a lead event
Meta ranks the identifiers it can match in three tiers. Email and click ID come first, then phone, country, external ID, browser ID, birthdate and Facebook login ID, and last the names, city and postcode. A lead form reaches the top tier almost by accident, because the email is typed into the form and the click brings the click ID with it.
| Identifier | Where it comes from on a lead page | Handling |
|---|---|---|
| the field the visitor fills in | hash it | |
Click ID (fbc) | the _fbc cookie, or assembled from the fbclid | send as it is |
| Phone | the form field, with the country prefix | hash it |
| External ID | your own reference for the lead | hash it |
Browser ID (fbp) | the cookie the Meta Pixel writes | send as it is |
| Country | a form field, or read from the phone prefix | hash a two-letter code |
| First and last name | the name fields, when the form has them | hash them |
| Client IP address | the request that carries the submission | send as it is |
| Client user agent | the headers of the same request | send as it is |
The bottom two are not in Meta’s priority list, which does not make them optional. A website event needs the user agent, Meta’s developer guidance counts the IP address among the high-quality parameters, and neither costs anything because the request already carries them.
How to clear the diagnostics, in order
- Send the Lead from your server with the form data, hashing email and phone. The score is calculated on Conversions API events, so this is where the gains are.
- Capture
fbcandfbpon the landing page visit, and keep them with the click. When there is no_fbccookie, buildfbcon the server from thefbclidin the landing page URL. That single change moves FBC coverage. - Pass the visitor’s real address and user agent, taken from the request that submits the form, never your own server’s or a proxy’s. If the landing page domain answers over IPv4 only, add IPv6 support to the server and the DNS, which is Meta’s own remedy for the IPv4 diagnostic.
- Add an
external_idwith your own lead or click ID. If the pixel sends one too, use the same format on both sides. - Leave empty fields out. A blank phone field means no phone parameter, not a placeholder string that every empty submission shares.
- Wait 48 hours and read the score of the Lead event itself, not of the PageView.
Normalise before you hash
Meta expects contact details to arrive already hashed, and it does not accept names, emails or phone numbers in clear text. Normalise first, then hash with SHA-256. The order matters more than the algorithm:
| Field | Raw input | Normalised | Rule |
|---|---|---|---|
[email protected] | [email protected] | spaces removed, lowercased | |
| Phone | +39 347 123 4567 | 393471234567 | digits kept, country prefix added |
| Country | Italy | it | two-letter code, lowercased |
| First name | Mario | mario | lowercased, punctuation removed |
Three failures account for most of the rest. A phone without the country code hashes into a different string from the same number with it, and Meta wants the prefix even when every lead comes from one country. A double-hashed email is a valid-looking string that matches nobody. And fbc, fbp, the IP address and the user agent are never hashed, so hashing them breaks the match silently.
What the API cannot tell you
A value normalised the wrong way does not raise a diagnostic. Its hash looks correct and simply does not match, which is why Meta’s reference of customer information parameters publishes the expected SHA-256 output for its examples: test your own code against those values before you blame the audience.
The report also says nothing about lead quality. Event Match Quality measures whether Meta can recognise the person behind an event, not whether that person was worth calling, and a perfect score would not make Meta’s reported lead count equal your own. That gap is covered in Meta Pixel, Conversions API and attribution.
Two practical rules follow. Keep sending a well-matched, deduplicated Lead, since duplicate events distort the same numbers from the other direction: pixel and Conversions API deduplication has the event ID rules. And keep making budget decisions on the leads your own system received, which is what Event Match Quality for lead generation argues in full.
DropCPL sends the Lead event to Meta from the server with an event_id for deduplication plus fbc and fbp, and counts the lead on your own landing page, so the CPL you decide on comes from the leads you actually received. The dataset those events land in is described in Meta dataset, and if you ever need to start over, how to delete a Meta dataset or Pixel covers what that costs you.