Self-reported attribution: how to use “How did you hear about us?” answers alongside your sales data

A “How did you hear about us?” question measures what the customer remembers starting it, not the click that brought them in. It is your only record of the friend, the truck and the podcast, and it cannot tell paid search from organic. Keep it as a second column beside tracked revenue. Never add the two together.

Every attribution argument eventually reaches the same person: the customer, who knows how they found you and would tell you if you asked. That is the case for self-reported attribution, and it is a good one. Tracking only sees the steps that left a trace in a system. A recommendation over a fence, a van parked on the next street, a podcast mention heard while driving, none of these produce a click, so none of them appear in any report built from clicks.

The case against it is just as simple. People are bad at recalling where they first heard of something, they do not know the difference between an ad and a search result, and they answer the question you asked in the order you asked it. A business that switches from tracking to a survey has traded one blind spot for another.

The useful position sits between the two: ask the question well, code the answers consistently, and read them against the lead source you can actually match to the sale. Where the two agree, you can stop arguing. Where they disagree, the disagreement is the finding.

What is self-reported attribution?

Self-reported attribution is asking customers where they heard about you and counting their answers by channel. It records the customer's memory of the cause, usually the first exposure that stuck, so it complements tracked attribution, which records the path the lead took into your systems.

It goes by several names: “How did you hear about us?” (HDYHAU in the trade), a source question, a post-purchase attribution survey. The mechanics are the same in each. A question appears on a form, on a booking call, on a checkout confirmation or on an invoice, and the answer is stored against the customer or the order.

What makes it different from every other method in the attribution models comparison is the source of the evidence. Pixels, call tracking, UTM parameters and CRM fields all observe behavior. A survey asks for testimony. Testimony is good at the things behavior misses and poor at the things behavior records precisely, which is why the two work best side by side rather than one replacing the other.

Why survey answers disagree with your tracking

Survey answers and tracked sources disagree because they measure different moments. Customers report the exposure they remember, often a person or a place, while tracking records the last measurable step, often a search for your name. Both can be true about the same sale.

Take a homeowner whose neighbor recommends a roofer. A week later they search the roofer's name, click the first result, which is a branded ad, and call. Tracking gives the sale to Google Ads. The customer says “my neighbor.” Neither is lying. The neighbor caused the sale; the ad captured it. This is the pattern behind branded search taking credit for demand it did not create, and it is the single most common disagreement you will see.

The wording of the question decides which of those moments you hear about.

For most businesses that already track their leads, the first-hear version is the more valuable one, because tracking already tells you the last step. A survey that asks for the last step again mostly confirms what you knew.

How to word the “How did you hear about us?” question

Ask one question, in the customer's words, with six to nine options and an Other box. Use “first hear” wording if tracking already covers the last step. Randomize the order on a written form, keep Other last, and never change the list mid-quarter.

• One question, not three. “How did you first hear about us?” on its own. A follow-up asking which ad, which station or which friend gets low completion and messy text.
• Options in the customer's vocabulary. “Google search”, not “paid search” or “SEO”; “Facebook or Instagram”, not “Meta”; “Saw our truck or a yard sign”, not “OOH”. Customers cannot classify their own exposure into your channel taxonomy.
• Six to nine options. Fewer, and everything lands in Other. More, and people pick the first plausible one without reading the rest.
• An Other box, always. It is where new channels show up first: a local Facebook group, a supplier, a radio host you did not know mentioned you.
• No “ads” option. Customers cannot reliably say whether a result was an ad, so an “online ad” answer is mostly noise. Let tracking answer paid versus organic.
• Fixed for at least a quarter. Changing the list changes the answers. Compare months asked the same way.

The order of the options matters more than most people expect, and in opposite directions depending on whether the customer reads the list or hears it.

That last pair of numbers is the risk in a dropdown. A short list does not just sort the answers; it suggests them. If “Google” is on the list and “a local Facebook group” is not, the group will be under-reported by a wide margin, and you will never know, because the people it reached picked the nearest option rather than typing in Other.

This matters for any business that books by phone. A receptionist who reads “Was it Google, Facebook, a friend or one of our trucks?” at the end of a booking call is asking a list question orally, at the end of a long conversation, with options that overlap. That is close to the worst case the study describes. Ask it as an open question instead, “How did you first hear about us?”, and let the receptionist pick the category from what the customer says.

Where to ask it, and what each placement misses

Ask where nearly every customer passes and the answer is stored against something you can match: the web form, the booking call, or the order confirmation. Each placement has a bias, so pick one per customer path and record which one produced each answer.

A business that sells mostly through calls and visits should ask at booking, because that is the moment every customer passes and the moment the answer can be written next to a phone number. An online store should ask on the confirmation page. Whatever the placement, the answer has to land on a record that carries the customer's phone number or email, or you will not be able to set it beside the tracked source later.

How to code free-text answers into channels

Map every answer, including Other text, to one fixed list of channels before counting anything. Write the mapping down as a table, apply it the same way every month, and keep “Don't remember” and blank as categories of their own rather than spreading them across the rest.

The referral row deserves its own care. A business with a referral program can tell a tracked referral (a code, a named referrer) from general word of mouth, and the two should stay apart; the referral revenue post covers how to count the tracked kind. The survey is what tells you how big the untracked kind is.

Worked example: 200 closed jobs, survey answer versus matched lead source

Cross-tabulate each closed sale's survey answer against the lead source it matched to. The diagonal is where the two agree; the off-diagonal cells show which channels are capturing demand that something else created, and which unmatched sales the survey can explain.

Here is a month at an invented home-services company: 200 closed jobs, each with the answer recorded at booking and the lead source it matched to by phone number. “Organic and Business Profile” covers unpaid search and the Google Business Profile listing; “Unmatched” is a sale with no lead in the file.

Read it row by row, and five things come out that neither source shows alone.

1. The survey cannot split search. Of 60 customers who said “Search”, 28 matched a Google Ads lead and 22 an organic or Business Profile lead. A survey-only report would give all 60 to one line, or split them by guesswork. Paid versus organic is a question for tracking.
2. Word of mouth is being captured by Google. Of 52 customers who named a person, 26 (50%) still arrived through a Google ad, an organic result or the Business Profile, almost certainly by searching the company's name. Tracking alone credits Google Ads with 14 of those sales. That does not make the ads worthless (a competitor bidding on your name would take some of those clicks), but it does mean cutting brand awareness to fund search would starve the source the search is harvesting.
3. Totals that agree can hide rows that don't. The survey says 24 social; tracking says 22 Meta Ads. A manager comparing the two totals would call it confirmed. The cross-tab shows only 15 customers are in both: 9 social answers matched another channel or nothing, and 7 Meta Ads sales came from customers who gave another answer or none. Compare per sale, never per total.
4. Offline brand has a size. 14 customers named the vans or yard signs, and 9 of them matched a search lead. No other measure in the business would ever have shown those 14.
5. The survey explains most of Direct / Unknown, but it does not prove anything. 49 sales matched no lead. Of those, 42 have an answer, led by word of mouth (20) and repeat customers (10). That is a far better account of unattributed revenue than a blank bucket, but it is testimony, so report it as a reason beside the bucket rather than moving the revenue into channels.

Two more figures from the same table are worth tracking monthly. The answer rate is 168 of 200, or 84%, which is high because staff asked at booking; a web field that is optional usually does far worse. And the agreement rate, on the two rows a customer can actually judge, is 65 of 77 matched sales: 50 of the 55 matched “Search” answers went to a search channel, and 15 of the 22 matched “Social” answers to Meta Ads, 84% in all. A sharp drop in either is a sign the question or the tracking has broken, and it usually shows before anything else does.

How to report it next to tracked revenue

Report self-reported answers as a second column beside tracked revenue by channel, never added to it. Use tracking for decisions about paid channels and the survey for decisions about brand, word of mouth and offline spend, the parts of the budget tracking cannot see.

• Two columns, one row per channel. Tracked revenue in one, sales by survey answer in the other. A total that adds them counts the 26 word-of-mouth-through-Google sales twice.
• Say which question each column answers. “Delivered the lead” for tracking, “customer says started it” for the survey. Readers otherwise assume both mean the same thing and pick whichever suits them.
• Never rewrite tracked credit with survey answers. Some tools let the survey answer override the click when they disagree. That turns a measured fact into a remembered one for exactly the sales where memory is least reliable.
• Keep the blank row visible. A survey report that quietly drops 32 non-answers looks cleaner and shows the remaining channels larger than they are.
• Use the gap to set budget questions, not answers. A channel with far more survey mentions than tracked sales is doing work upstream; a channel with far more tracked sales than mentions is mostly capturing. Both are worth testing before moving money.

For a business just starting, the small-business attribution guide sets out how the survey fits into a monthly routine, and lead source tracking covers making sure the tracked column exists in the first place. The survey adds the most to a business whose tracking is already solid; it cannot rescue a lead file with no phone numbers in it.

CloseRev produces the tracked side of this comparison. It reads two CSV files you export, closed sales and leads, matches each sale to leads on normalized phone and email, and by default credits it to the source of the earliest matched lead; last touch is an option. Only high-confidence matches count automatically, medium- and low-confidence ones stay reviewable, and sales with no match stay in Direct / Unknown. It does not read survey answers, so the cross-tab described here stays a spreadsheet step.

When self-reported attribution misleads

Self-reported attribution misleads when the question is optional, when staff answer for the customer, when the list leaves out the channels that matter, and when its answers are allowed to overwrite tracked credit. Each failure looks like a confident number.

• Staff fill it in. When the field is required and the customer was not asked, the default option wins. A sudden jump in the first option on the list is the tell.
• Long consideration. A customer who researched for three months remembers the last thing, or the most vivid, not the first. Long sales cycles need the tracked first touch more than the survey.
• Repeat customers counted as new. “Used you before” belongs outside acquisition numbers entirely, or customer acquisition cost is understated.
• A list that suggests the answer. As the Pew experiment shows, a short list under-reports anything not on it. Read Other every month and add an option when a write-in passes a few percent.

None of these are reasons to drop the question. They are reasons to keep it in its lane: the customer's account of how it started, written next to the record of how the lead arrived, and read for the places where the two tell different stories.