Free Plan SaaS: Why the Wrong Users Find You First

A free plan SaaS strategy can look like the safest possible way to grow: remove the price barrier, collect signups, prove value, and convert engaged users later. But a recent founder postmortem from video-hosting startup Stillhaven shows the harder truth: when the offer, acquisition channel, and product risk model are misaligned, free users may create costs without ever becoming prospective customers.

The important lesson is not that every startup should kill its free tier. It is that a free plan does not attract “the market.” It attracts a very specific slice of the market, shaped by the words in your ads, the jobs your onboarding permits, and the friction you remove at signup.

The free-plan experiment: 35 accounts, zero upgrades

Stillhaven, a small Michigan-based video-hosting company, recently shared a candid account of closing its free plan to new users after roughly three months. The experiment produced 35 free accounts, no upgrades to paid plans, and five suspensions for terms-of-service violations. Most accounts reportedly uploaded a large file quickly, then did not return after their first day or two.

That is a small sample, and it would be a mistake to treat it as a universal conversion benchmark. Yet the post is valuable precisely because it publishes the kind of uncomfortable operational data many founders keep private: a zero in the upgrade column, moderation work that did not show up in initial unit economics, and a meaningful gap between the company’s intended use case and the use case users inferred from its marketing.

The founder describes accounts uploading apparent copyright-infringing material, pornography, and files distributed through external links. Automated scanning helped identify some adult content, but it could not reliably identify pirated movies or television episodes from video frames alone. Instead, the team relied on behavioral and contextual signals such as filenames, rapid uploads after signup, repeated uploads, and suspicious sharing patterns.

Stillhaven ultimately replaced open free signup with a card-required 14-day trial and a $9-per-month entry plan. Existing legitimate free users were moved to the entry plan at no charge. That change does more than add revenue potential: it changes the qualification system before storage, streaming, moderation, and support costs accumulate.

The real problem was not “free”—it was user intent

The easiest interpretation of this story is that free plans are bad. The more useful interpretation is that the free plan attracted people with a different job to be done than the product was built to serve.

Stillhaven originated from a practical workflow: hosting video for clients to review, comment on, and approve, particularly for professional creative work such as aerial and drone footage. A legitimate customer in that workflow is not shopping for unlimited anonymous uploads. They are trying to move a project through review, gather feedback, protect an unreleased cut, and deliver an approval outcome.

Those are radically different motivations.

A person searching for “private video hosting” may mean “I need a password-protected review link for a client.” Another may mean “I want a host that will not ask questions about the material I upload.” The identical word—private—can carry incompatible commercial intent.

That is why pricing alone cannot fix acquisition. A free plan gives the public a low-friction route into your product, but your acquisition message determines which public arrives. If the message signals anonymity, unlimited storage, unmoderated distribution, or an easy workaround for established platforms’ rules, the product can become attractive to precisely the people least likely to become healthy customers.

The founder’s conclusion is especially sharp: the issue was not a loophole in the plan but who the plan was effectively built for. Limits, scanner rules, and disposable-email blocks can reduce certain forms of abuse. They cannot transform an audience with the wrong intent into a qualified customer base.

Why Google Ads keyword matching changed the audience

Stillhaven spent $401.48 on Google Ads over 30 days, generating 120 clicks and three free-plan signups. The resulting cost was about $3.35 per click, close to the founder’s estimate of $3.30. The most expensive ad group was built around “private video hosting,” but its actual search terms reportedly included phrases such as “anonymous video host” and “upload video anonymously.”

This is the acquisition failure at the center of the postmortem. The business thought it was paying to reach people with a confidential professional workflow. It was also reaching people whose primary requirement was concealment.

Google Ads itself documents that positive keywords can match searches that are similar to the keywords in an account, rather than only exact copies of those keywords. The search terms report exists specifically so advertisers can review queries that triggered ads and identify irrelevant terms to exclude. Google also notes that negative keywords behave differently from positive keyword matching and do not automatically cover all close variants or semantic expansions. (support.google.com)

“Private” is a benefit, not a market segment

Founders often build campaigns around feature language because it seems straightforward. Terms such as private, secure, fast, unlimited, encrypted, anonymous, simple, and affordable all sound customer-friendly. But they are broad descriptors, not necessarily precise indicators of the buyer’s workflow.

For a video-hosting platform, “private” could point to at least five different needs:

• A production company sharing an unreleased cut with a client.
• A marketing team sending a campaign preview for approval.
• An educator sharing a lecture with enrolled students.
• A company distributing an internal training video.
• A person seeking a place to upload material without accountability.

The first four can be strong business use cases. The fifth may impose disproportionate legal, moderation, payment, and infrastructure risk. If all five searches land on the same ad and the same “upload now” landing page, the conversion funnel has already failed its qualification role.

Search-term review is not optional maintenance

The practical lesson is not merely “add more negative keywords.” It is to treat search-term analysis as a core product-discovery activity. Search queries reveal how people classify your product, what outcomes they expect, and which meanings your positioning accidentally invites.

For a workflow-focused video platform, better keyword clusters might describe the work rather than the infrastructure:

1. Client video review and approval.
2. Video feedback for agencies.
3. Share rough cut with client.
4. Password-protected video review.
5. Branded video portal for customer approvals.

Those searches are narrower and may have lower volume. That can be an advantage. A smaller pool of people actively trying to complete the workflow your product solves is usually more valuable than a large pool attracted by a generic feature phrase.

A free plan SaaS offer is also a risk policy

Every product tier makes an implicit statement about what behavior is allowed before trust has been established. For low-cost software with negligible marginal usage, that may be mostly a financial question. For video, email, file storage, communications, payments, marketplaces, and AI generation tools, it is also an abuse-prevention question.

Video hosting has variable costs that are easy to underestimate. Storage costs increase with uploads. Transcoding consumes compute. Playback creates delivery and request volume. Investigating suspicious material consumes human attention. A single abusive account may also create exposure with payment processors, cloud providers, copyright holders, and legitimate customers who expect a reliable service.

The same pattern applies elsewhere:

• A free email API can attract spam attempts and sender-reputation damage.
• A free file-sharing product can attract malware or infringing uploads.
• A free AI image tool can attract prohibited-content generation and automated scraping.
• A free communications API can attract fraud, phishing, or account-takeover campaigns.
• A free marketplace listing product can attract fake inventory and payment scams.

In each case, the headline infrastructure cost may be manageable. The second-order costs are often what change the math: trust and safety reviews, blocked payment accounts, support load, incident response, deliverability problems, and the opportunity cost of a small team constantly handling bad-fit users.

The Stillhaven team had already introduced sensible controls: blocking disposable email domains, treating Gmail dot and plus variants as one inbox, flagging suspicious filenames, reviewing flagged frames, and imposing bandwidth and monthly upload caps. Those controls are useful. But their limited impact makes an important distinction clear: fraud prevention tools handle abusive behavior; positioning and qualification prevent much of the bad-fit demand from arriving in the first place.

The most useful activation metric may be a return action

One community response to the founder’s post proposed a better signal than signup volume: whether a free user comes back in week two and plays their own upload. That may sound minor, but it captures a critical product truth.

A person who uploads a file and leaves may be using the service as disposable infrastructure. A person who returns to review a video, checks comments, shares an updated cut, or manages access is more likely to be completing an ongoing workflow. The latter user has evidence of recurring value—and therefore a plausible reason to pay.

For a video-review product, a stronger activation event might be something like:

A new account uploads a video, creates a controlled share link, receives or leaves feedback, and returns within seven days to review activity or upload another version.

That definition is much more revealing than “account created” or even “first upload completed.” It combines setup, sharing, collaboration, and repeat engagement.

Measure qualified activation, not raw activation

Every SaaS product should define a version of qualified activation. The question is not simply whether the user touched the product. It is whether they performed the sequence that proves they are using it for the intended job.

For example:

The exact events differ, but the principle is stable: qualified activation should correlate with a durable workflow and a credible path to paid use. If it does not, growth metrics can disguise a demand-quality problem.

ChatGPT referrals are a new acquisition channel—and a new qualification problem

One surprising detail in Stillhaven’s account was a cluster of five signups attributed to ChatGPT referrals in a 12-hour period. The founder concluded that the site’s AI-readable content had worked as intended, but was apparently surfaced to users seeking free private hosting for questionable purposes.

This does not mean ChatGPT “caused” abusive users or that companies should hide from AI search. It means AI referrals, like search-engine referrals, reflect the language and intent in a user’s prompt. If a product is positioned around phrases that map to high-risk demand, an AI assistant may surface it when those phrases appear in relevant queries.

OpenAI says publishers who allow OAI-SearchBot can track inbound traffic from ChatGPT search results, including through the utm_source=chatgpt.com parameter. OpenAI also distinguishes OAI-SearchBot, which supports search visibility, from GPTBot, which relates to potential training use. (help.openai.com)

That makes AI referral traffic measurable, which is good news for operators. But measurement needs to go beyond sessions and signups.

What to track for AI-referred traffic

Founders should create a separate source segment for ChatGPT and other AI assistants where possible, then compare it with organic search, paid search, direct, partner, and referral traffic. Look at downstream quality, not just top-of-funnel volume.

Useful questions include:

• What landing pages are AI-referred visitors reaching?
• Which phrases appear in the content on those pages?
• Do AI-referred users reach qualified activation at the same rate as other users?
• Do they create support tickets, abuse alerts, or chargeback risk at a higher rate?
• Do they select the right product tier and use the expected features?
• What query intent might explain a sudden cluster of signups?

The answer may be to change how a page frames the product, not to block bots or remove all AI-readable content. A page that says “free anonymous video uploads” will attract a very different audience than one that says “client video review, feedback, approvals, and controlled sharing.” Both describe video hosting. Only one describes a professional workflow.

Why card-required trials can be healthier than free forever

Stillhaven’s replacement strategy is a 14-day free trial with a card on file, paired with a low-cost $9 monthly entry tier and smaller trial limits. This approach adds friction, but friction is not automatically bad. Good friction screens for seriousness when a product has meaningful variable costs or abuse exposure.

A card requirement is not a perfect identity system. Fraudulent cards, virtual cards, and determined attackers still exist. Nor should founders assume that requiring a card will solve positioning errors. If ads still target people searching for anonymity, the product may continue attracting unsuitable traffic.

Still, a card-required trial can improve the economics and quality of the funnel in several ways:

1. It creates a commitment signal. Users willing to start a trial with a payment method are more likely to be evaluating a real solution rather than looking for disposable capacity.
2. It limits repeated account creation. Payment credentials add another deterrent beyond email address checks.
3. It supports a clear conversion path. The trial naturally moves into a paid plan unless the customer cancels, subject to transparent consent and reminder practices.
4. It makes usage limits easier to explain. A trial can be designed to demonstrate the workflow without becoming an indefinite no-cost hosting account.
5. It protects operational capacity. Smaller upload and bandwidth allowances keep a trial from becoming a low-friction bulk-storage product.

The trade-off is lower signup volume. That is acceptable if the lost signups were never qualified prospects. A funnel should not be optimized for the number of accounts it can create; it should be optimized for the number of sustainable customers it can serve.

Build onboarding around the job, not the upload button

Another community comment highlighted an overlooked issue: if the first experience tells users to “upload anything,” a free plan can become a file-dumping invitation. The intended customers, by contrast, arrive with context: a project, a client, a review deadline, collaborators, and a particular outcome.

That suggests a better onboarding pattern. Ask users what they are trying to accomplish, then configure the first-run experience around that answer.

For a professional video platform, onboarding could begin with a lightweight question such as: “What are you sharing?” Options might include client review, agency approval, internal training, portfolio delivery, event footage, or something else. A second question could ask who needs access: one client, a team, external reviewers, or a public audience.

Those answers can improve both activation and risk management. Someone selecting “client review” can be guided to create a password-protected link, request feedback, and add a branded project title. Someone who refuses to provide context, immediately uploads a massive file, or selects a workflow inconsistent with their behavior may warrant more scrutiny.

Qualification should be proportional

Do not turn signup into an interrogation. Excessive questions reduce conversion and can create privacy concerns. The goal is not to verify every user’s entire story before they see the product. It is to request enough context to make the intended use case clear, route good users into the right workflow, and identify obvious mismatches early.

A practical progressive approach is:

• At signup: ask for role, intended workflow, and country or business type if relevant.
• Before high-cost actions: require email verification, a payment method, or an additional confirmation.
• After suspicious patterns: apply review queues, reduced limits, or manual approval.
• For high-risk verticals: use explicit allowlists, contracts, or sales-assisted onboarding.

This makes trust a product capability rather than an afterthought handled entirely by support.

The infrastructure debate matters, but it cannot rescue a bad funnel

The Reddit discussion also moved into infrastructure economics. The founder said Stillhaven currently runs on AWS and is evaluating other platforms including Cloudflare. One commenter noted Cloudflare R2’s lack of egress charges; the founder replied that AWS CloudFront’s newer flat-rate plans can make delivery costs predictable at a smaller scale, though request limits are especially important for HLS streaming because it generates many small requests.

Both points can be true. Cloudflare R2 currently advertises no internet egress charges, but bills for storage and operations; its published pricing includes 10 GB-month of free standard storage, one million Class A operations, and 10 million Class B operations per month. (developers.cloudflare.com) AWS’s CloudFront flat-rate plans, introduced in late 2025, bundle CDN delivery with services such as WAF, DDoS protections, Route 53, logs, edge compute, and S3 storage credits for a fixed monthly price with stated usage allowances and no overage charges. (aws.amazon.com)

For founders, the strategic point is bigger than AWS versus Cloudflare: cheap bandwidth does not make abusive demand good demand.

A lower egress bill can reduce the direct cost of a bad account. It does not eliminate moderation time, payment-provider risk, copyright complaints, support queues, brand damage, or the risk that abusive behavior crowds out legitimate users. Infrastructure optimization is valuable once the company has a healthy definition of who should be using the product.

Model requests, not only gigabytes

This is particularly relevant for streaming products. HLS and similar adaptive streaming approaches break video into segments, meaning a single viewing session can generate numerous requests. A platform evaluating “free egress” or a bundled data-transfer allowance should also model request volume, storage duration, transcoding, cache behavior, logs, and security tooling.

A useful free-tier cost model should include:

• Average stored GB per activated user.
• Average upload and transcoding workload.
• Average delivery GB and requests per viewer.
• Percentage of users who share externally.
• Percentage requiring manual review.
• Support minutes per account.
• Payment, fraud, and compliance costs.
• Expected conversion rate among users who reach qualified activation.

Without those inputs, “the free tier only costs a few dollars per user” can become a dangerously incomplete calculation.

A framework for deciding whether to keep a free tier

Founders do not need to choose between unlimited freemium and no trial at all. There are several models, and the right choice depends on marginal costs, product-led growth potential, sales motion, abuse risk, and the speed at which customers recognize value.

Keep a free plan when these conditions are true

A continuing free tier can make sense when:

• The product has low marginal cost per active user.
• The free experience naturally encourages team invitations, sharing, or ecosystem adoption.
• Free users can receive meaningful value without consuming the core paid resource indefinitely.
• Abuse can be managed with low-friction controls.
• Conversion pathways are clear and measurable.
• Free users generate strategic value even if they never pay, such as templates, marketplace supply, product awareness, or referral loops.

Examples may include lightweight design tools, note-taking products, developer utilities, collaboration software with limited usage, and products where the free version itself creates network effects.

Prefer a trial or paid entry plan when these conditions are true

A card-required trial, usage credit, or low-cost paid entry plan is usually stronger when:

• Every use creates storage, compute, delivery, or third-party API costs.
• The product can be abused anonymously or at scale.
• The key value can be demonstrated within days or weeks.
• A serious prospect has a clear business reason to evaluate the product.
• The product requires hands-on onboarding, moderation, or compliance review.
• The company is small enough that bad-fit accounts distract directly from serving paying customers.

Stillhaven’s move fits this second category. The new model gives prospective buyers a way to test the actual product, while limiting the chance that the service becomes a destination for disposable uploads.

What founders should do before changing pricing

Before removing a free tier, do not focus only on conversion percentage. Examine the complete cohort journey by channel, use case, and behavior. A zero-conversion cohort from one search campaign does not prove that a free plan cannot work. It proves that this particular offer acquired through this particular funnel did not work.

Use this diagnostic checklist:

1. Segment by acquisition source. Compare paid search, organic search, AI referrals, communities, partnerships, and direct traffic.
2. Inspect query intent. Review actual search terms and landing-page promises, not only campaign keywords.
3. Define qualified activation. Identify the actions that show a user has begun the intended workflow.
4. Measure return behavior. Look for second-week usage, repeat projects, teammate invitations, and meaningful content management.
5. Calculate fully loaded costs. Include moderation, support, infra, fraud, compliance, and founder time.
6. Review abuse by cohort. Determine whether suspicious behavior clusters around a channel, geography, landing page, or signup flow.
7. Test a smaller commitment. Try a usage-limited trial, paid starter tier, credit card verification, or manual review for expensive actions.
8. Rewrite the promise. Lead with the specific outcome for the ideal customer rather than generic feature language.

The result may be a retained free tier with better limits. It may be a trial. It may be a self-serve paid plan. Or it may be a sales-assisted motion for high-value users. The objective is not to preserve an ideology about freemium; it is to construct a repeatable path from acquisition to successful customer use.

The bigger lesson: acquisition channels are product decisions

Stillhaven’s story is a useful warning for builders in AI, SaaS, cloud infrastructure, and digital services. Marketing is not just what happens before the product. The people your messaging recruits influence the support burden, feature roadmap, content-moderation needs, infrastructure profile, and perceived reliability of the product itself.

A free plan SaaS strategy is therefore a compound decision involving pricing, positioning, product design, operations, abuse prevention, and measurement. The offer tells people what they can get. The language tells them whether it is for them. The onboarding tells them what to do next. The limits tell them what behavior is economically viable for your business.

Stillhaven did not merely learn that 35 people would not pay. It learned that an open offer combined with ambiguous acquisition language had pulled in a population that did not resemble its existing customers. Moving to a limited, card-required trial is an attempt to align the evaluation experience with the product’s intended professional workflow.

That is the mindset founders should take away: do not ask whether free users are good or bad. Ask whether the users your free offer attracts can realistically become successful customers—and whether the business can safely support everyone else.

FAQ

Should every SaaS company offer a free plan?

No. A free plan works best when marginal costs are low, abuse risk is manageable, and free users can reach value without consuming expensive resources forever. Products with meaningful storage, compute, delivery, moderation, or compliance costs may be better served by a trial or low-cost entry plan.

What is the best metric for a free plan SaaS product?

Track qualified activation rather than signup count. The best metric is a product-specific action sequence that demonstrates the user is completing the intended workflow and returning for continued value, such as collaborating, sending, sharing, reviewing, or managing a second project.

Does a card-required trial stop abuse?

It reduces some low-effort abuse and creates a stronger commitment signal, but it is not a complete solution. It should be combined with clear positioning, sensible usage limits, email verification, behavioral monitoring, and review processes for high-risk activity.

How can founders prevent bad Google Ads traffic?

Review the search terms report frequently, add negative keywords for irrelevant terms, separate campaigns by workflow and buyer intent, and write landing pages around a specific job to be done. Do not rely on feature words such as “private” or “secure” without clarifying the professional use case they support.

Should companies block AI search referrals if they attract poor-fit users?

Usually, no. First measure AI-referred traffic separately and identify the pages and messages attracting the mismatch. AI referrals can be valuable when the content clearly describes the intended buyer, workflow, limits, and product category. The goal is better qualification, not less discoverability.