Maintenance
Part of How to test discovery platforms for creator coverage
Discovery platforms mistakes and the habits that fix them
Discovery platforms mistakes rarely start in the query. These are the sort-order, follower-band and empty-result habits that quietly choose your shortlist.
Most of the damage a search tool does happens in the ten minutes after the results load. The query itself was fine.
What follows is a series of small, fast decisions that decide who gets contacted. They are the discovery platforms mistakes that rarely get logged: which column sorted the list, which filter felt safe, what went to a client, and what nobody wrote down.
These are the mistakes worth fixing, because none of them requires a different product.
The platforms these mistakes apply to
The habits below apply across the discovery platforms agencies actually use. CreatorIQ, Traackr and Upfluence are suite-style platforms built around a large influencer database. Modash, Klear, HypeAuditor and Heepsy are search-first discovery and analytics tools. Aspire combines a creator marketplace with campaign management. The mistakes are not unique to one vendor; they are how the default sort, follower bands and empty results behave in all of them.
What to take away
- The default sort is a decision somebody else made for you, and it is usually popularity.
- Contactable is not interested. A list of routes into people is not a pipeline.
- An empty result is ambiguous. It can mean nobody is doing this, or it can mean the index cannot see them.
Letting the sort order pick the shortlist
Results arrive in an order. Almost nobody changes it, and the top of the list gets read carefully while the bottom gets scrolled past. Whatever the tool sorts by therefore becomes your selection criterion, and it is generally some version of size or engagement rate.
Fix this by re-sorting deliberately before reading anything, or by reading the list backwards once. Every platform named here has a sort control; use it before the first read. The names at the bottom of a popularity sort are where the underpriced candidates sit, and the first pass through a list is the only one anyone gives full attention.
Filtering by follower band because the filter is there
Size bands are the easiest filter to apply and among the least informative. They act as a proxy for price, badly, and they exclude exactly the people who are cheap relative to their influence in a niche.
If your real constraint is budget, get quotes and let refusals do the filtering. If your real constraint is reach, say so and be honest that you are buying reach, which is a different campaign from the one most briefs describe. A constraint like that belongs in the search spec rather than in a filter setting.
Six quick habits and what each one produces
Six quick habits
| The habit | What it produces |
|---|---|
| Accepting the tool's category taxonomy as your category | A shortlist shaped by somebody else's tagging decisions, missing everyone who sits between two labels |
| Reusing last quarter's shortlist without rechecking | Contacting people who have changed direction, gone quiet, signed elsewhere, or died in the index |
| Sending the raw export to a client | A conversation about follower counts, which is the least useful comparison available |
| Reading silence from an agency as disinterest | Losing a candidate who was simply in a queue behind other enquiries |
| Adding a candidate because they are famous enough to please someone internally | A booking made for internal reasons that the audience did not ask for |
| Recording no reason for a rejection | Re-shortlisting the same wrong person next quarter, and arguing about it again |
The agency row is worth expanding. Represented creators reply on a different clock, because the enquiry sits in a queue with commercial priorities attached, and a chase in a week is normal rather than pushy. What that queue looks like from the other side is set out in the guide to how representation works.
Treating the shortlist as a booking list
A shortlist ranked by preference invites booking down it in order until the budget is gone. That produces a campaign of near-identical creators, because a single ranking is generated by a single criterion.
Group by role instead: one person explains the product, one makes it look ordinary in daily use, and one reaches an audience no one else on the list touches. Use the list or tag feature in the platform to keep the groups separate, then book across the groups.
A tool cannot make this campaign design decision, which is the step most often skipped between search and outreach.
Confusing an index gap with a market gap
You search a category, get almost nothing, and call the space empty. Sometimes that is true, and exciting when it is.
More often it is the streetlight effect: you found what the index could show you and read that as the market's shape. The index may miss the platform those creators use, the language they post in, or hold only people who registered with the vendor.
Check before you act on it. Search the same thing by hand on the platform itself, ask one person who works in that market, or run the query in a second product such as Modash or HypeAuditor.
The guide to how discovery data is built explains the structural reasons an index cannot see certain creators. Those are unrelated to how big the vendor says the index is.
Judging the product on a search designed to succeed
The demo search works because whoever chose it knew the answer. The same instinct ruins internal trials: people test the tool on the brief they have already solved.
Test on the brief you failed at, then run a search that produced nothing useful last quarter. Record what the tool cannot do, not what it can, with a date, because interfaces and coverage change and a six-month-old impression is not evidence.
The documents that keep a trial honest are the shortlist columns and the rejection log.
Mistaking a score for a finding
Every product offers some quality or authenticity signal. HypeAuditor and Modash are two widely used examples; their scores are built from engagement patterns, follower growth and audience demographics. Most such signals are a form of anomaly detection, which finds the unusual rather than the wrong. Where those scores fail in both directions is covered in the guide to verification tooling.
Treated as a sorting aid a score is useful. Treated as a verdict it produces two expensive errors: dropping a good candidate on a number nobody can explain, and repeating that number to a client as though it were a measurement.
Keep the doubt and the evidence separate. A doubt is a reason to ask for the creator's own platform analytics for a stated period.
When the tool is not the problem
If shortlists keep failing, the brief is the usual cause. A brief that names an audience but not a tone, or that hides a constraint until the shortlist arrives, will defeat any index.
The other common cause is buying the wrong thing entirely. Where the work is transactional and the relationship does not matter, matching and paperwork handled by a market may be a better fit than searching. TikTok Creator Marketplace, YouTube BrandConnect and Aspire are examples of that route, and those trade-offs are set out in the guide to creator marketplaces. The four guides referenced in this article are 'How talent representation works', 'How discovery data is built', 'Verification tooling' and 'Creator marketplaces'.
Common questions
How often should a shortlist be refreshed?
Treat anything older than one campaign cycle as a list of leads rather than a shortlist. Rechecking is quick, and the facts that change most are availability, representation and recent competitor work.
Should we tell candidates they came from a search tool?
It makes no difference to them and it is not worth a sentence. What does matter is that the first message shows you have actually watched their work, which is the difference between a reply and silence.
Our client insists on a follower minimum. What do we do?
Meet it and also present two candidates below it, with the evidence for each. Arguing the principle rarely works; a specific person with a specific post attached sometimes does.
Is it worth running two tools in parallel?
Only after you have checked the overlap. Export the same query from both, match on creator handle, and count the unique names in each list. Two indexes that return substantially the same people are one index sold twice, and the second subscription buys an interface.
