How NexSpark matching works

NexSpark matching works in two stages: an AI engine scores every identity-verified member in your area against your answers, in both directions, and a human matchmaker reviews each proposed pairing before it's sent. Nothing is auto-sent.

By Matthew Firth, founder of NexSpark · Oct 5, 2026 · 6 min read

Illustration of a couple at a sunlit kitchen table, each holding a card and comparing their lists over a shared open notebook, flat editorial style in navy and coral

This page is the plain version of what happens between the moment you finish the questionnaire and the moment a match report lands in your inbox. It's written for people who want to know what a machine is deciding about them and what a person is deciding, because at NexSpark those are two different jobs and we think you should be able to see the seam.

What the AI does: Searches every verified member in your area, scores each pair in both directions, enforces dealbreakers as rules, and produces a ranked shortlist.
What a person does: Reads both profiles, decides whether the introduction is worth your evening, and writes the report explaining why.
What's rejected outright: Any pair where a dealbreaker conflicts, either person is outside the other's age or distance range, or either person isn't verified.
Response window: Five days to accept or decline an introduction. No penalty for declining.
Verification: Identity and age verified through a third-party provider before anyone can be matched. No criminal background checks.
Cost: Free during launch. After launch, $599 once per person, refundable in full until a connection opens. Details on the pricing page.

Stage one: what the AI scores

Your questionnaire is your profile. There's no bio to write and no photo to lead with; the answers themselves, about your values, how you live day to day, how you connect, how you bond, and what you can't live with, are what get compared.

The engine starts by pulling every member who is verified, is seeking what you're seeking and vice versa, and whose overall pattern of answers sits closest to yours. That first cut is a similarity search across your whole questionnaire, not a filter on any single answer. From there it scores each pair two ways and blends them:

Overall similarity. How close the full shape of your answers is to theirs. This is the larger share of the score.
Answer-by-answer comparison. Each attribute is compared with the rule that fits it. Some are scored for agreement, like faith and the life you're building toward. Some are scored for near agreement, so "wants kids" against "open to kids" earns half credit rather than none. Some are scored for a moderate difference, because the research on temperament and social energy says a little contrast helps; "your calm balances their spark" is a rule in the scorer, not just a line in the marketing. And attachment style is scored as a pairing: a secure partner buffers almost anyone, while anxious with avoidant is marked down as a specific combination, not because the two are "different."

Every attribute also carries the importance you gave it. If you said faith matters a great deal and they said it barely registers, the score reflects both of those, not an average.

Then the part most matching products skip: the score is computed in both directions, from you to them and from them to you, and the pair is ranked by the weaker of the two. A match one person would love and the other would decline isn't a strong match, so it isn't ranked like one.

What's a rule, not a preference

Before any of that scoring happens, a short list of things removes a pair entirely. These aren't weighed against the good parts of a match. They end the comparison.

Dealbreakers, in both directions. A dealbreaker you set is enforced against them, and theirs is enforced against you. If the other person's answer on your dealbreaker is unknown, they're excluded rather than let through on a guess.
Age range, mutually. You have to be inside their range and they have to be inside yours.
Distance, mutually. Same rule. Each of you has to be within the other's travel range, measured from the ZIP code you gave us, which is never shown to anyone. There's also a hard ceiling no preference can override, so nobody gets introduced to someone they could never realistically meet.
Verification. A member who hasn't completed identity and age verification, or whose date of birth and location aren't confirmed, isn't in the pool at all. A paid introduction can't rest on a guess about either.

Where we're careful: the similarity search pulls the closest few hundred members before the rules above run, so in a small city the shortlist can be short. We'd rather send you nothing than pad it. The engine also won't re-run for you more than once a day, which is a cost decision we're telling you about rather than hiding.

Stage two: what a human reviews

The engine's output is a ranked shortlist of candidates for you. That's where its authority ends. It doesn't send anything, and it can't.

A human matchmaker (today, that's me) opens the shortlist and reads both profiles side by side, with the score breakdown showing where the pair agrees, where it complements, and where it merely tolerates. The question at that point isn't "is the number high" but "would I put these two people at a table together and expect them both to be glad they came." Pairs that pass on numbers and fail that question get set aside. Pairs that pass both get a written match report: who this person is, in alias, and the specific reasons we think you fit, in plain sentences rather than a percentage.

That report is what reaches you. Not a profile to browse, not a feed, one introduction with the reasoning attached.

What you see, and how long you have

Your match report names the other person only by an alias, shows that they're identity and age verified, and tells you why we paired you. You have five days to accept or decline. Declining costs nothing and counts against nothing; we'd rather hear a no early than watch a maybe expire.

If you both accept, that's when we introduce you properly and names are revealed. If one of you recorded an optional short intro video, the other sees it only after a mutual yes, and our team reviews every video before a match can see it. Early conversation happens inside NexSpark, where it can be moderated, until you both decide to take it elsewhere.

How long it takes

We don't promise a turnaround, because the honest answer depends on who's verified in your city this month. The engine refreshes your candidates no more than once a day; the review step is a person reading, and a match report goes out when a pairing is worth sending, not on a schedule. During launch we're matching in Atlanta, and the other cities we're building toward are listed openly.

What this isn't

It isn't auto-matching. No introduction is generated and delivered by the model alone. The Trust Center states this as policy, and this page explains how it's true in practice.
It isn't a background check. Verification confirms identity and age. We don't run criminal background checks or screen against offender registries, and our safety disclosure says so without hedging.
It isn't surveillance. We don't scan your social accounts or buy data about you. The questionnaire is the whole input.
It isn't a subscription. Free during launch; afterward, $599 once per person, covering twelve months of introductions, refundable in full until a connection opens.

If you want the research behind the scoring choices, the science page cites it, including the parts that are inconvenient for a matchmaking company. If you want the step-by-step of using the product, the help guide walks through it with screenshots.

Common questions

What does NexSpark's AI actually score?
Two things, for every pair and in both directions: how similar your overall questionnaire answers are, and how your individual answers compare on values, faith, children, lifestyle, temperament and attachment style, weighted by how much each of you said it matters. Some answers are scored for agreement, some for a moderate difference, and attachment styles are scored as specific pairings rather than as same-or-different.

Does a human review every match?
Yes. The AI produces a ranked shortlist of candidates for each member and that is where its job ends. A human matchmaker reads both profiles, decides whether the introduction is worth sending, and writes the match report that explains why. Nothing is auto-sent.

What happens if a dealbreaker conflicts?
The pairing is removed before scoring. Dealbreakers are enforced as rules in both directions, not treated as preferences to trade off, and a candidate whose answer on your dealbreaker is unknown is excluded rather than given the benefit of the doubt.

How long do I have to respond to an introduction?
Five days. There is no penalty for declining, and names are revealed only when you have both accepted.

Does NexSpark run background checks?
No. Every member completes identity and age verification through a third-party provider before they can be matched, which confirms they are who they say and at least 18. NexSpark does not run criminal background checks or screen against offender registries, and says so plainly in its safety disclosure.

See who the engine shortlists for you

Answer the questionnaire, get your archetype report instantly. The match report comes only after a person has reviewed the pairing.

Reveal your ideal match