How to read design agency reviews: a worksheet for startup founders
Stars tell a founder less than they seem to. This guide shows what review platforms verify, why agencies' review profiles lean positive, and how to turn each review into a worksheet row and a question.
Illustrative situation: a seed-stage founder has two product design agencies on the shortlist to redesign an onboarding flow. Agency A shows 4.9 stars from 31 reviews on its Clutch profile. Agency B shows 4.6 stars from 9 reviews, plus a few LinkedIn recommendations for its lead designer. The co-founder says the stars settle it. The founder suspects that 31 reviews about website projects may tell her less about her project than 9 reviews about product work, but she has no method for saying so. The scene is invented.
Humbleteam publishes this guide and is a digital product design agency with reviews and testimonials of its own, so every check below applies to ours too. The full pre-hire process, with briefs, proposals and reference calls, is in our guide to evaluating design agencies as a startup founder. This guide covers one input to that process: what to do with the reviews you find. Where it mentions US law, it is general information and not legal advice.
What does each review source check, and what does it leave to you?
Each platform publishes its own rules. The table below summarizes what those pages say, as read on 2026-10-26.
| Source | Who can post | What the platform says it checks | What it leaves to you |
|---|---|---|---|
| Clutch | A current or former client, one review per client company, for a project in progress or finished within two years | Login through LinkedIn, Google or a company email, a software check of the reviewer’s digital footprint, and editor review. Outcome: published with a Verified badge, published without one, or not published | The check compares project details with what the agency publishes about its own work, so it confirms plausibility more than results. Reviewers may be anonymous |
| G2 | Reviewers who validate through LinkedIn, a business email, or a personal email plus a screenshot | Human moderation, a conflict-of-interest check, and labels such as Validated Reviewer, Current User and Incentivized | G2 describes itself as a guide for software purchases, says reviews are subjective experiences and not expert opinions, and says it has not confirmed reviewers’ qualifications. Check what is being reviewed |
| Google Maps | People using Maps | A policy that bans paid, incentivized, conflict-of-interest and fake reviews, enforced by machines and human reviewers that look at the review, the account and unusual activity at the place | The pages we read do not describe checking that the reviewer hired the business |
| LinkedIn recommendations | A connection of the person recommended | The recipient accepts a recommendation and can hide it later | The recipient decides what shows. The pages we read do not describe checking that the two worked together. Public profiles show a count and at most two recommendations |
| The agency’s own site | Whoever the agency chooses | No platform sits between you and the agency | Selection, editing and truth are all up to the agency, and you cannot verify them from the page |
Clutch’s methodology page says every review is read by an editor, and that in 2025 Clutch rejected 10% of submitted reviews for signs of falsified information. That is Clutch’s own figure. It also shows why a platform check is a filter, not a guarantee. Only verified reviews count toward a Clutch company’s review score, so a Not Verified review is visible but carries no weight in the rank. Read the badge before the stars.
Position matters too. By Clutch’s description, directories sort sponsored listings first, then Verified providers, then everyone else by Clutch Rank, and Verified status needs a paid subscription or an advertising product that includes it. Clutch’s help center adds that phone-interview reviews are available only to clients of providers on those paid plans. A top spot in a directory partly reflects what the agency pays, and how a review was collected can depend on its plan.
Why are the stars friendlier than the full client list?
Agencies choose whom to ask. Clutch tells providers they can invite clients to leave a review by email, in person or through badges on their website, using a customized review link. The pages we read do not say which clients a provider must invite. Other platforms do spell it out. Trustpilot’s business guidelines list inviting only people known to have had a positive experience as something a business must not do, and FTC staff’s guide for review platforms says not to ask for reviews only from people you think will leave positive ones. A rule against a practice shows the practice is possible. So ask the agency how it chooses whom to invite.
Research on consumer reviews points the same way. Nan Hu, Paul Pavlou and Jie Zhang argue that online ratings carry two self-selection biases: people who buy a product tend to like it, and people with extreme opinions are more likely to write. In their experiment, where everyone reviewed the same music CD, ratings were bell-shaped and lower on average than the same CD’s Amazon ratings. A study of more than 280 million reviews from 25 platforms found most platforms show strongly polarized ratings, and that the self-selection behind this makes the average rating less informative. Both studies cover consumer products and services. We found no equivalent study of design agency reviews, so treat the direction as a warning, not a measurement.
The same logic applies to the other sources. A LinkedIn recommendation appears only if the recipient accepts it. A testimonial on an agency’s site is there because the agency put it there. FTC staff say the consumer reviews rule does not ban selective use of positive reviews in marketing, though non-representative ones can be deceptive under the FTC Act. A flattering page is normal, so read it for what is missing.
What does the FTC rule on fake reviews cover?
The FTC announced its final rule on August 14, 2024, and 16 CFR Part 465 took effect on October 21, 2024. Per the FTC’s press release, it prohibits fake or false reviews and testimonials, including AI-generated ones, and buying reviews that must express a particular sentiment. It also prohibits insider reviews that hide the writer’s connection to the business, company-controlled sites that falsely claim to give independent reviews, review suppression through groundless legal threats or intimidation, and buying fake social media followers. Courts may impose civil penalties for knowing violations.
Three limits matter for a reader, per the FTC’s questions and answers and the rule text.
- A platform that only hosts reviews is exempt, and it has no duty under the rule to investigate them. A business that publishes testimonials on its own site is distributing them and can be liable if they are fake.
- The rule has no specific ban on asking only happy customers for reviews. The FTC says the practice could still violate the FTC Act and points to the Endorsement Guides (16 CFR Part 255). Section 255.2(d) says advertisers should not, when procuring, suppressing, boosting, organizing or publishing reviews, distort what consumers think of their products.
- The rule is written about consumer reviews. It defines a consumer as a person who purchased, used or otherwise had experience with a product, service or business. We found no passage in the rule text, the press release or the Q&A that says whether a company’s review of an agency counts. A lawyer would have to answer that.
So the rule works better as a list of red flags than as a guarantee. It does not certify any review you read. The red flags it names are insider reviews with no disclosure, paid sentiment, an unusually large number of reviews in a very short period (an example the FTC gives of a sign that purchased reviews are likely fake), and a “best agencies” list run by one of the agencies and presented as independent. That last one applies to rankings on any site, ours included, so check who runs the page and whether it says so.
How do you read one review?
Fill this in for every review you rely on, before you look at the average. The “dates” row follows Clutch’s own two-year window. The rest is our proposal.
| Check | Look for | Weak signal | Question it gives you |
|---|---|---|---|
| Dates | Project dates and publish date, both within about two years | An old project with a new review, or a burst of reviews in a short stretch | Is the team that did this the team I would get? |
| Label | Verified badge on Clutch, Validated Reviewer or Current User on G2, any incentive note | Not Verified, an incentive, or an anonymous reviewer | Can you introduce me to this client? |
| Reviewer | Job function and company details; whether this person worked with the team week to week | A title far from the work, or no detail | What was your role on the project day to day? |
| Scope | Services listed, deliverables named, what was handed over | “Great design work” with no scope | Which parts did the agency own, and which did your team do? |
| Match to you | A similar product type, stage, team size and engineering support | A branding or marketing-site project when you need product design | Which parts would not carry over to a product like mine? |
| Cost and timeline | Dates, cost if shown, and the scheduling and cost sub-ratings, not only quality | Cost hidden, dates vague, high quality but low scheduling | Did scope, price or dates change, and who asked for the change? |
| Specifics | A named decision, an awkward moment, a number with its source | Praise that could describe any agency | Tell me about a disagreement and how it was settled |
| Criticism | Answers to “areas for improvement”, 3-star and 4-star reviews, the agency’s reply | Only 5-star reviews, or a defensive reply | Did the agency fix what you flagged? |
| People | Names or roles of who did the work | No mention of the team | Who was on your project, and for how much of their time? |
| Pattern | Spread of dates and ratings, repeated phrasing, a reviewer with a link to the agency | Five-star reviews arriving together, identical wording, an investor, partner or former employee as reviewer | How do you choose whom to ask for reviews? |
| What is missing | Reviews for your project type, from the last year, from work that ran into trouble | None for your type, or none recent | Of your last ten clients like me, how many reviewed you? |
Clutch’s reviews follow a fixed format: background, challenge, solution, results and ratings, with separate questions on areas for improvement. The format makes the “Criticism” and “Cost and timeline” rows easy to fill. Clutch also defines a conflict of interest as a reviewer who works or worked for the provider, has a family tie, or holds a financial stake, so the “Pattern” row is partly a check on whether the platform caught something.
The last row matters most. A profile can be accurate and still say nothing about the work you need, because you cannot see the clients who were never asked.
What happens when you compare two agencies this way?
This example continues the invented scene. The agencies and every figure are made up.
| Item | Agency A | Agency B |
|---|---|---|
| Clutch rating and count | 4.9 from 31 reviews | 4.6 from 9 reviews |
| Ratings | 27 at 5 stars, 4 at 4 stars | 6 at 5 stars, 2 at 4 stars, 1 at 3 stars |
| Verification | 27 Verified, 4 not | All 9 Verified |
| Timing | 19 of the 31 posted in the last four months | Spread over two years, 2 older than 18 months |
| Reviewer roles | Mostly marketing managers | Founders and heads of product |
| Project types | Mostly website redesigns | 6 product design, 3 website |
| Criticism | One line in four reviews, none detailed | The 3-star review says handoff to engineering ran three weeks late |
| Agency reply | None | A specific reply to the 3-star review |
| None checked | Two recommendations for the lead designer, one from a former client’s CTO |
The star gap points to A. The worksheet points somewhere else. A’s strength is volume, but 19 of 31 reviews in four months, mostly from website projects, say little about an onboarding redesign. The founder cannot yet tell whether the burst follows a push for reviews, a new service line or something else. B’s evidence matches the project type but is thin: nine reviews and one detailed complaint. Neither agency is out. Each gets its own to-do list, and the lines in the reviews become questions.
| Review line | Question |
|---|---|
| A: “Great communication throughout” | Which meetings did you attend each week, and who from the agency joined them? |
| A: 19 reviews in four months | How do you choose whom to invite, and what prompted this round of requests? |
| B: “Handoff ran three weeks late” | To the agency: what changed after that project? To the reviewer: what did your engineers need that was missing? |
| B: nine reviews in two years | Which other clients of the last 24 months in my project type did not review you? |
The decision moves from “which has more stars” to “which gap can be closed before signing.” A has to show product design references from the last 12 months. B has to show that the handoff problem was fixed and that the designer on that project is available.
What can reviews never tell you?
Reviews describe a past engagement with a past team. They come from one person’s view, and G2 says reviews are subjective, may be incomplete, and rest on qualifications it has not confirmed. Clutch’s relationship check compares a review with the agency’s own description of its work. That supports plausibility but does not measure whether the redesign improved anything. The self-selection research above is a reason not to lean on the average alone.
A review is evidence for your questions, not for the decision itself. Whatever a review cannot answer is what a conversation with a named client, or a small paid trial, has to cover.
What can you produce now?
Three things before the next agency call. One worksheet per shortlisted agency, filled in for each review you plan to rely on. Five questions per agency, taken from the lines in its own reviews. And a short note stating which agency’s review evidence is thin for your project type and what would close the gap. With those in hand, the star average becomes one line in the note and stops being the verdict.
