Flashpoint.AIFlashpoint.AIdocs

Send us traffic

The fastest way to field with us. No API work on your side — you receive an entry link, you send respondents to it, and we return them to you with an outcome. Most partners are live within the hour.

You do not call our API for this. Everything below is a link contract: one URL you receive, five URLs you provide.

What you need to provide

Two things: some commercial basics, and five return URLs. No API work either side.

Commercial basics

Send these once, with your first enquiry:

MarketsCountries and languages you can field
SpecialismsAny sector or audience you're genuinely strong in — B2B, healthcare professionals, IT decision-makers, automotive, finance, parents, low-incidence medical. This is often the reason we'd route a study to you rather than to an exchange, so don't undersell it
CPIYour cost per complete, by market. By LOI and incidence band if your pricing varies that way
CapacityRealistic completes per day, per market
Panel provenanceWhere your respondents come from — proprietary panel, router, affiliate traffic, client list. We don't object to any of these, but we price and screen differently, and finding out later is a problem
Respondent idsFormat, and whether an id is stable for the life of the panelist or just the session
DeduplicationWhether you dedupe within a study, and across studies
ContactA named person for live fieldwork issues, with hours
DPAYour data processing agreement

Pricing on this path is agreed per project rather than quoted through an API, so the CPI you give us is the figure we hold for that study. Tell us when it changes.

Five return URLs

OutcomeWhat it meansRequired
completeFinished and qualifiedYes
disqualifyFailed a screener questionYes
terminateEnded by survey logic after screeningYes
over_quotaQualified, but their quota cell was fullYes
bad_qualityBlocked by fraud or duplicate detection before startingStrongly recommended

If your protocol doesn't separate disqualify from terminate, give us the same URL twice — we'd rather you were explicit than have us guess.

bad_quality is the one partners most often skip, and it's the one that costs you money. It's where we return a respondent that Research Defender blocked at the door — a bot, a VPN, a repeat device. Without it they fall back to your terminate leg, which means you can't tell "we rejected this person as fraudulent" from "they failed a screener", and you can't replace them. Supply it.

Send the five to your Flashpoint contact, or the buyer can add them directly in the app. Once saved, your source is reusable across every study the team runs with you.

The entry link

You'll receive a link shaped like this:

https://flashpointopinions.com/a1b2c3d4-5678-90ab-cdef-1234567890ab?rid=<RESPONDENT_ID>&fp_src=f0e1d2c3-b4a5-6789-0123-456789abcdef
PartWhat it isWho sets it
/a1b2c3d4-…The surveyWe do
ridYour respondent identifierYou substitute, per respondent
fp_srcIdentifies you as the sourceWe do — send it back unchanged

Two rules:

Always substitute a real rid. It is how a respondent is attributed to you, how we return them to the right place, and how reconciliation works later. A link shipped with rid= left empty loses every respondent's identity.

Never strip fp_src. It tells us which partner the respondent came from, which decides which return leg they get. A respondent arriving without it on a study with more than one source cannot be reliably attributed, and we will not guess — a guess sends someone to the wrong vendor's passback.

You may append any other query parameters you like. We capture all of them and store them against the response, so anything you need back at reconciliation time — your session id, supplier id, campaign — survives the round trip.

Returning a respondent

When the respondent finishes, we redirect the browser to the URL you gave us for that outcome.

Your URLs can carry placeholders, which we substitute at redirect time:

TokenSubstituted with
{{rid}}The respondent id you sent on entry
{{response_id}}Our internal response id, useful for support tickets
{{QUESTION_LABEL}}That respondent's answer to any question, by its label

Values are URL-encoded on substitution. An unknown token, or one with no value for this respondent, becomes an empty string rather than breaking the URL.

Example. You provide:

complete     https://panel.example.com/return?pid={{rid}}&status=complete
disqualify   https://panel.example.com/return?pid={{rid}}&status=screenout
terminate    https://panel.example.com/return?pid={{rid}}&status=screenout
over_quota   https://panel.example.com/return?pid={{rid}}&status=quotafull
bad_quality  https://panel.example.com/return?pid={{rid}}&status=security

A respondent who entered with rid=XYZ789 and completed is redirected to:

https://panel.example.com/return?pid=XYZ789&status=complete

Piping an answer back works the same way — if your side wants the respondent's self-reported age for your own profiling, and the survey has a question labelled AGE:

https://panel.example.com/return?pid={{rid}}&age={{AGE}}

URLs must be absolute and begin with http:// or https://. Relative paths are rejected when the source is saved.

Our quality bar

Worth reading before you field rather than after, because it's the thing most likely to end a partnership.

Every respondent is screened twice. Research Defender runs before they reach a question — fraud, duplicates, VPNs, repeat devices. ReDem scores open-ended answers after the fact. Respondents who fail either are excluded from the deliverable.

Excluded respondents aren't billed. You won't pay for traffic we reject, and we won't pay for it either. We'll send you the respondent ids so you can look at them on your side.

We watch source-level patterns, not just individuals. The signals that get a source paused for review:

  • Open-ended answers that are templated or near-identical across respondents
  • A language profile that doesn't match the market — a notably high share of English answers in a non-English market is the usual tell
  • Quality scores clustered at a single value rather than distributed
  • Completion times tightly grouped in a way organic traffic isn't

Any one of these can be innocent. Together they mean we stop the source, review, and come to you with the evidence before resuming. It's recoverable, and it goes much better when the expectation was set in advance.

Replacement. Excluded completes don't count toward quota, so a study will keep running to make up the shortfall. If exclusions run high enough that you can't fill, we'd rather talk about it early than let the study stall.

Test before you field

Ask your Flashpoint contact for a test link against a draft study. Preview responses are flagged and excluded from quotas, completes and analytics, so you can walk the flow as many times as you need without touching live data.

Walk all five legs and confirm each one arrives on your side with the outcome you expect:

  • Complete the survey → lands on your complete URL
  • Fail a screener → disqualify
  • Trigger a mid-survey terminate → terminate
  • Enter against a filled quota cell → over_quota
  • If you can simulate a blocked entrant → bad_quality

Check the rid that comes back is byte-for-byte the one you sent.

Go-live checklist

  • All five return URLs supplied and saved
  • All five legs walked on a test link, each landing where expected
  • rid round-trips unchanged
  • fp_src present on every link you'll field, and passed through untouched
  • Your side distinguishes over-quota from screen-out
  • Commercial terms and DPA in place
  • A named contact on each side for live fieldwork issues

If something looks wrong

Respondents arrive but aren't attributed to us. Almost always a missing or altered fp_src. Check the link you're actually fielding, not the one you were sent — rewriting through a router is the usual culprit.

Respondents reach the end and don't come back to you. The study has no return leg for that outcome, or the source wasn't attached before fielding started. Tell us the response_id if you have it.

Everyone comes back as a terminate. You likely didn't supply over_quota or bad_quality, so both fall back to the terminate leg.

Your completes don't match ours. Ours count only respondents who finished and qualified; a fingerprint count of everyone who touched the link will always be higher. Ask us for the field report, which breaks it down by outcome.

Reference

Entry parameters

ParameterRequiredNotes
ridYesYour respondent id. panel_id and PID are also accepted and normalised.
fp_srcYesYour source id. Pass through unchanged.
anything elseNoCaptured and stored against the response.

Return legs: complete, disqualify, terminate, over_quota, bad_quality

Template tokens: {{rid}}, {{response_id}}, {{ANY_QUESTION_LABEL}}

Next steps