An influencer post-purchase survey can show which creators customers remember when tracking links do not explain the purchase. Keep those answers beside your link records, match them to orders where possible, and report the overlap. A survey answer records a customer's recollection. It cannot establish every earlier interaction or prove that a creator caused the order.
Modash's measurement guide recommends surveys alongside links and codes, while acknowledging missing responses and memory errors. The useful output is a comparison between those sources, with unanswered surveys left visible.
Decide what you want customers to remember
Separate these two questions before choosing survey software:
- Where did the customer first discover the brand?
- What helped the customer decide to buy this time?
A creator may introduce a brand months before a customer buys through an email. Both answers can describe the same purchase without agreeing on one source.
Analytics also assigns credit according to a chosen model. Google's attribution documentation explains that models use rules or algorithms to distribute credit. Its paid and organic last-click model credits the last eligible channel, ignoring direct visits unless the whole path is direct. A first-discovery survey asks a different question.
Write down your survey's purpose and keep the discovery and purchase answers in separate fields. For link records, retain the attribution model and window used in the export. Choose the reporting window before reading results so a late purchase does not change category merely to improve the report.
Use a neutral question set
The following is a recommended starting questionnaire, not a validated instrument or a platform requirement. Show it after the order completes. Keep it optional and introduce its purpose with: "Help us understand how you found us. You can skip any question."
| Question | Response format | Handling rule |
|---|---|---|
| Where do you remember first hearing about us? | Short free-text answer; separate "I don't remember" choice | Record discovery in the customer's own words. |
| What, if anything, helped you decide to place this order? | Short free-text answer; separate "Nothing in particular" and "I don't remember" choices | Keep purchase reasons separate from discovery. |
| Can you name the account, person, or website you mentioned? | Optional free text, shown only when an earlier answer names a source without identifying it | Accept an incomplete answer. Do not suggest campaign creators. |
Ask both broad questions before requesting a creator name. Pew Research Center explains that earlier questions can introduce ideas people then repeat in open-ended answers. Opening with "Which influencer sent you?" assumes an influencer was involved and supplies the answer category you hope to measure.
If free text creates too much work, replace the first question with a single-choice version. Possible categories are a social post or video, a search result, a website or publication, a personal recommendation, an offline encounter, and an ad outside social media. Include "Other," "More than one source," and "I don't remember."
Test these labels with customers before launch. A social post could come from a brand, a creator, or an ad. Keep that answer as "social, source unknown" unless the optional detail identifies more. Do not automatically code every social answer as creator discovery.
Keep raw answers as well as your coded categories. Document how you handle misspellings, unknown handles, multiple names, and answers such as "Instagram." When a reviewer cannot identify a creator, retain the uncertainty.
Account for three sources of bias
Recall bias
Customers can forget an earlier encounter or confuse its timing. The UK Government Analysis Function's questionnaire guidance identifies elapsed time and how memorable an event was as factors in recall error.
Asking after checkout keeps the purchase decision recent. It does little to shorten the time since a much earlier discovery. Preserve "I don't remember" as a valid answer. Do not force a creator selection or reinterpret uncertainty as "direct."
Response bias and nonresponse
People who complete the survey may differ from people who skip it. For example, enthusiastic creator followers could be more willing to name their source. That is a possible bias to investigate, not a finding about your customers.
AAPOR's survey guidance warns that low participation can leave some types of respondents missing. A larger number of answers does not, by itself, resolve that problem.
Track eligible orders, orders shown the invitation, and answers received. Compare participation across new and returning customers or other existing order categories. Keep skipped questions separate from explicit "I don't remember" answers. If you change placement or offer an incentive, record the change before comparing periods.
Option-order bias
In a self-completed survey, people can favor choices near the top. Pew describes this as a primacy effect. If you use unordered source categories, randomize those categories where the tool permits it. Keep utility choices such as "Other" and "I don't remember" in consistent positions.
Randomization distributes order effects; it does not remove every bias. If your tool cannot randomize choices, document the fixed order. Avoid moving "creator" to the top during a campaign and treating the resulting increase as evidence of better performance.
Reconcile answers without counting an order twice
Here is a hypothetical monthly dataset. It contains 1,000 eligible completed orders and one survey invitation per order. There are 300 responses. An order counts as creator-mentioned if either broad answer identifies a creator under the documented coding rules.
The link column means the store's existing tracking rule assigned the order to a creator link. It is an order-level flag for this example, not a claim about a particular analytics export.
| Survey result | Creator-link order | No creator-link order | Total |
|---|---|---|---|
| Creator mentioned | 25 | 35 | 60 |
| Answered, no creator mentioned | 15 | 225 | 240 |
| No survey response | 80 | 620 | 700 |
| Total | 120 | 880 | 1,000 |
The calculations are reproducible:
- Response coverage among eligible orders is 300 / 1,000 = 30%.
- Creator mentions among responding orders are 60 / 300 = 20%.
- The overlap is 25 orders.
- Orders with either recorded signal are 120 + 60 - 25 = 155.
- Survey-only creator mentions account for 35 orders.
Report "155 orders with at least one creator signal," with the breakdown. Do not report 180 by adding the two totals. Do not call all 155 incremental orders or treat the 35 survey-only mentions as verified missed link conversions.
Likewise, multiplying 20% by all 1,000 orders produces 200, but the calculation assumes respondents represent nonrespondents. This dataset gives no basis for that assumption.
Keep both creator identities when a survey names one creator and the link record names another. An order-level overlap does not establish that the identities match. Review those disagreements separately. If you cannot join responses to orders, report aggregate results side by side and state that overlap is unknown.
Use disagreement to choose the next check
A rise in survey-only mentions gives you a reason to inspect creator names, timing, and tracking setup. Use the missing-traffic diagnostic to investigate link problems before declaring that every unmatched answer represents a tracking failure.
These answers also cannot tell you what buyers would have done without the campaign. Keep that question separate when you plan a measure of incremental impact.
Before the next campaign, test the questions with a few customers and ask them to explain what each question means. Then prepare the reconciliation table with explicit rows for unanswered surveys and unknown sources. Agree on those labels before the first response arrives.
