How to Source UGC for Ads From Existing Customers
How to source UGC for ads from customers you already have: four signals that identify who to ask, and the request that returns a usable file.
Sourced UGC is defined as customer-made content obtained by asking identified existing customers directly, rather than collecting whatever appears publicly or briefing a paid creator. Shoppers value it: 70% say they trust opinions that other consumers post online [Nielsen, 2012]. To source UGC for ads, find the right people with four signals, then make one concrete request for a specific photograph and take advertising permission in the same message. You will learn the four signals and the wording of the request.
- Use four signals to find contributors: repeat purchase, long review, recovered issue and high score.
- Ask for one specific photograph, because an open invitation to submit returns nothing.
- Name paid advertising in the permission sentence, in the same message.
- Do not improve the file, because the unpolished quality is the whole asset.
- Send twenty targeted asks instead of a broadcast, because they beat it on response rate and on content mix.
The best advertising creative most stores will run this year is already sitting in their customer base, unasked for. Omniconvert has measured social proof and creative performance across the CROBenchmark dataset of 7,000+ websites in 15+ industries, against 248+ audit criteria, over 13 years in eCommerce [CROBenchmark Report 2026, Omniconvert]. Willingness to supply customer content is high, but requests are rare, vague or sent to the wrong people. Omniconvert Commander (formerly Nexus) is the AI for eCommerce growth engine that unifies commerce data, prioritises experiments by True Profit and generates campaigns and creative you approve before they go live, and Brandfeel by Omniconvert, its sentiment-intelligence layer, is the part that reads review and survey language closely enough to tell you which customers are worth asking. Last updated: October 2026.
This article covers sourcing. For the loop that keeps it running, see UGC flywheel, for the collection system underneath it build a Review-Generation system, and for the no-spend version get customer UGC without a big budget.
Why asking the whole list fails, and targeting works
Concentrating on twenty likely contributors changes the mathematics completely. These are people with a demonstrated relationship to the product, who frequently already have a usable photograph, and who are being asked something specific by a brand they have chosen more than once. The response rate on a well-targeted ask of this kind is in a different class from a broadcast, and the content that returns is better because it comes from people who actually use the thing.
The four signals that identify a contributor
A contributor signal is defined as a data point that predicts a customer's willingness to supply content. Each signal predicts willingness for a different reason, which is why combining them works better than any one alone.
Repeat purchase of the same item. The strongest signal available, and the most underused. Somebody who bought the same product twice has told you it worked, with money, which no survey score matches. Pull this list first.
A long unprompted review. Length and specificity are the signal here rather than the rating. A person who wrote four sentences naming particular details, unasked, is a person who wants to talk about the product. The ask is not a new behaviour for them, it is a smaller one.
A support issue that ended well. Counterintuitive and reliable. A customer whose problem was handled generously has a story with a resolution in it, and is frequently more willing to help than someone with no history at all. The condition is that the resolution was genuinely good, which your support records already know.
A high survey score. The weakest of the four on its own, because a score is cheap to give. It is useful as a second signal: a nine out of ten who also bought twice is a near-certain yes.
| Signal | What it predicts | Where the data sits | Strength alone |
|---|---|---|---|
| Repeat purchase, same item | The product genuinely worked | Order history | Strongest |
| Long unprompted review | Willingness to talk about it | Review platform | Strong |
| Support issue resolved well | A story with a resolution | Support tickets | Moderate, often surprising |
| High survey score | General goodwill | NPS or post-purchase survey | Weak alone, good as a second |
| Any two signals together | A near-certain yes | All of the above, joined | The actual shortlist |
| No signal at all | Nothing, and asking costs list value | The rest of the list | Do not ask |
The request that makes a customer source UGC for ads
The wording does more work than the targeting, and the common version of it fails for a diagnosable reason.
An invitation to share your experience or submit content hands the customer an open brief. They now have to decide what you want, judge whether their photograph is good enough, and guess at the format. Most people resolve that uncertainty by doing nothing, which is a rational response to an ambiguous request.
A specific request removes all three decisions. Ask for a photograph of the product in the room they use it in. Say it can be taken on a phone and does not need to be good. Say where it may appear. That message has one action in it and the action is small.
Two further elements raise the return rate noticeably. Say what it is for in plain terms, because people help with a stated purpose more readily than with a vague one. And ask one short question alongside the photograph, such as what they would tell someone considering it, since the answer frequently becomes the advertising copy that runs beside the image.
Then take the rights in the same message. A permission sentence naming product pages, paid advertising, email and social, written in ordinary words, costs you one line and saves a chase that otherwise happens weeks later when somebody is building the advertisement.
Selecting and using it without ruining it
A customer photograph arrives slightly crooked, lit by a kitchen window, with a cluttered background and the product not quite centred. Everything in a brand team's training says to fix those things. Fixing them produces a competent brand photograph of a product in a home, which the audience has seen ten thousand times and discounts accordingly.
Select for the same property. Between two files, prefer the one with more specific context over the one that is technically better. Nielsen's long-running work on advertising trust has repeatedly placed peer recommendation above brand-authored messaging, and the Baymard Institute's product-page research has found shoppers actively looking for evidence from other buyers at the point of decision. Both findings depend on the content reading as genuinely from a customer, which polishing is the fastest way to destroy.
What a marketer should do this week
- Run the two-signal query. Repeat buyers of one product who also wrote a review of any length. This is usually a list of a few dozen and it takes one query.
- Pick twenty and write one message. One specific photograph, in the place they use it, taken on a phone, with the purpose stated and one short question attached.
- Put the permission sentence in that message. Product pages, paid advertising, email, social. Plain words. Keep the reply with the file.
- Place the first three files unimproved. On the product page they belong to. Crop to fit the slot and change nothing else.
- Send a thank you afterwards, with a small credit. Unannounced, after the fact. This is what produces the second contribution and the referral.
Where your order history, review text, support tickets and survey responses sit in four separate systems and have to be joined by hand before the two-signal query is even possible, that unification is the problem Brandfeel and Omniconvert Commander address together, with every proposed campaign staying something a person approves before it goes live.
FAQ: sourcing UGC for ads
How do you source UGC for ads from existing customers?
Identify the right people, then make one concrete request. Four signals find the people: repeat purchase of the same item, a review written at length without prompting, a support issue that ended well, and a high survey score. Then ask for one specific photograph of the product where they use it, rather than inviting a submission, and take advertising permission in the same message. A specific request to a likely person returns a file. An open invitation to a list returns almost nothing.
What makes customer content work in a paid ad?
Specificity and the absence of production. The pieces that perform tend to show the product in a real setting with context around it, at an angle no art director would choose, with the kind of detail a brand would tidy away. Resist improving them. Re-lighting, straightening and cropping a customer photograph into brand-standard composition removes the one quality it had, which is the visible evidence that a real person took it.
What permission do you need to use it in advertising?
Written permission that names paid advertising explicitly, collected in the same message as the request. A general consent to use a photograph is not a consent to spend money promoting it, and the distinction matters both legally and in the relationship. Name the four placements in plain words, product pages, paid advertising, email and social, keep the reply, and store it with the file so nobody has to ask twice.
Should you offer an incentive?
A small token after the fact works better than an offer made up front. An incentive attached to the request changes who replies, pulling in people motivated by the reward rather than by the product, and it shifts the content toward what they think you want. A discount or a credit sent afterwards, unannounced, reads as thanks rather than payment and tends to produce a second contribution later.
How many customers do you need to ask?
Start with twenty well-chosen people rather than a broadcast to your whole list. A targeted ask to customers who match the four signals converts at a rate that makes twenty enough to produce several usable files, and it protects the list from an invitation most recipients will ignore. Broadcasting the same request to thousands produces a worse response rate, a worse content mix, and a list slightly more tired than it was.
Teams that source UGC for ads often send the request to the whole list and a few dozen people reply. The reply rate looks poor, so the team concludes customers will not make content. The people most likely to say yes were never singled out, and the open invitation gave everyone else three decisions to make.
AliveCor used Omniconvert to run a structured A/B testing programme and achieved +21% conversion rate, +5% revenue per visitor, and 94% statistical relevance across their experiments. [Omniconvert, AliveCor case study] Test sourced assets the same way: run them against your current creative before you scale the spend.
Ask twenty likely customers for one specific photograph
The content you want is already owned by people who have already chosen you twice, and the reason you do not have it is that nobody asked them specifically. So run the query: repeat buyers of one product who also left a review. Take twenty names. Send one short message asking for one photograph of the product in the room they use it in, taken on a phone, with the purpose stated, one question attached, and a permission sentence that names paid advertising by name. Then place what comes back without improving it, and send a quiet thank you afterwards. The whole sequence is an afternoon of work against a list you already own, and it reliably produces better advertising creative than the same afternoon spent briefing anybody.