Beyond the Feature Factory: Finding What Users Want
A familiar SaaS scenario plays out every quarter. Sales wants the deal-saving feature. Support wants fewer tickets. Leadership wants a competitive response. Meanwhile, buyers are asking ChatGPT, Claude, Perplexity, and Google AI which product to choose before your team ever gets a chance to explain the roadmap.
Product discovery has changed because the buying journey has changed. Teams still need to confirm that a problem matters, that a workflow is painful enough to fix, and that users will switch behavior. They also need to know how AI systems describe the category, which sources they cite, and whether the product shows up in the right prompts. For SaaS teams, discovery now sits closer to research, positioning, and distribution than it did even two years ago.
That shift creates a practical requirement. Discovery has to cover user need and AI visibility at the same time. Teams that treat those as separate tracks usually miss obvious signals. A prompt test can expose weak positioning. Citation analysis can show which proof points the market trusts. Query expansion methods like query fan-out analysis for AI search discovery can reveal adjacent problems buyers ask about before they use your category language.
I use the same core rules I trusted before AI search became part of the funnel. Start with evidence. Check behavior before stated preference. Compare interview themes against product usage, search intent, win-loss notes, and support friction. Then pressure-test what you found in the environments buyers now use to evaluate software.
That is why this guide pairs established discovery methods with newer, AI-centric ones. It is written for SaaS teams that need better product decisions and better market visibility, especially teams using tools like MyMentions to track prompts, citations, and mention patterns alongside standard research inputs. If your team is also working on go-to-market for AI visibility, these techniques help connect discovery work to what prospects see when they ask AI for recommendations.
Table of Contents
- 1. AI-Powered Search Analysis & Prompt Testing
- 2. Customer Interview & Demand Validation
- 3. Search Intent Analysis & Keyword Research
- 4. Citation Source Analysis & Content Gap Identification
- 5. Competitive Benchmarking & Share of Voice Analysis
- 6. User Behavior Analytics & Sentiment Tracking
- 7. Stakeholder Feedback & Internal Alignment
- 8. Content Strategy & Positioning Refinement
- 9. Partner & Community Ecosystem Development
- 10. Conversion Optimization & Feedback Loop Implementation
- Product Discovery Techniques: 10-Point Comparison
- From Discovery to Delivery Your Action Plan
1. AI-Powered Search Analysis & Prompt Testing
A lot of SaaS teams still treat AI visibility like a brand issue. It's a discovery issue. If a buyer asks an assistant for the best tool for a job and your product never appears, or appears with the wrong positioning, you've learned something important about market perception.
AI assistants now shape validation before users ever touch your product. A recent write-up on Teresa Torres' LinkedIn points to a growing blind spot: prompt-level discovery is missing from many teams' process even as buyers rely more on AI answers and citations in product research, as discussed in this piece on AI visibility in product discovery.
Start with buyer-intent prompts, not vanity prompts. "Best CRM for a two-person consulting firm" tells you more than "What is Company X?"
A good prompt set includes use-case prompts, comparison prompts, alternative prompts, migration prompts, and pain-point prompts. Test them across ChatGPT, Claude, Perplexity, Copilot, and Google AI, because each system weights context and sources differently.
How to run prompt testing like discovery work
Use the same rigor you'd use in customer research. Keep the prompts stable for a period, log rank and framing, and review which citations keep showing up.
For teams doing this repeatedly, MyMentions has a useful explanation of query fan-out behavior in AI search, which helps explain why similar prompts can produce noticeably different product mentions.
Here are the patterns worth tracking:
- High-intent prompts: Ask the questions buyers ask right before shortlist creation.
- Prompt variations: Test phrasing your customers would use, not internal category language.
- Citation quality: Check whether the cited pages support the answer clearly or create ambiguity.
- Competitor framing: Compare whether rivals are presented as safer, simpler, more established, or more specialized.
Practical rule: Don't optimize for being mentioned once. Optimize for being recommended consistently across prompt variations.
If you need a strategic lens for this area, this guide to go-to-market for AI visibility is a useful companion to classic discovery work.
A short walkthrough helps if your team is new to this format:
2. Customer Interview & Demand Validation
A SaaS team ships the feature customers asked for, launches it, and adoption stalls. The interviews were real. The conclusion was wrong.
That usually happens when interviews collect opinions instead of evidence. Good discovery interviews focus on a recent incident: what triggered the search, what the buyer tried first, who joined the decision, what nearly killed the purchase, and what they used as a workaround. Demand validation starts there.
What strong interviews actually look like
Ask for the last time, not the ideal future state. "Walk me through the last time you had to onboard a new teammate into this workflow" gets you sequence, constraints, and stakes. "Would better onboarding matter to you?" gets you polite agreement.
In B2B SaaS, I use interviews to validate both demand and language. Buyers reveal the actual job, the internal politics around budget, and the wording they use in search, in Slack, and in procurement reviews. That wording should shape positioning, onboarding copy, sales enablement, and AI-facing content. It also pairs well with AI-powered keyword research when you want to compare stated pain with actual market phrasing.
A few practices raise signal quality fast:
- Anchor every answer in a specific event: Ask what happened first, what they tried next, and where the process broke.
- Trace the alternative path: Find out whether they used spreadsheets, internal docs, agencies, or a competing product before they considered your category.
- Capture exact language: Save phrases verbatim, especially complaints, success criteria, and comparison terms.
- Test willingness to change: Ask what made the problem expensive enough to solve now, not someday.
- Separate user pain from buyer risk: The end user may want speed. The buyer may care more about control, compliance, or rollout effort.
For modern discovery teams, interviews should also inform AI-era validation. If several prospects describe the problem one way but AI systems summarize your product another way, positioning drift is already happening. Reviewing how sentiment analysis for AI mentions works helps connect interview evidence with how the market and AI assistants frame your product in public.
The useful insight is rarely "customers want feature X." It is "they are trying to remove risk, cut handoff time, or replace a fragile workaround, and they will only switch if the change feels credible."
3. Search Intent Analysis & Keyword Research
Search intent analysis sits between customer research and positioning. It shows what buyers are trying to accomplish before they know your product name, and it exposes where your category language doesn't match theirs.
This is especially useful when the same product can be framed multiple ways. A team collaboration product might be searched as project management, internal wiki, product ops, knowledge base, or meeting notes software. If your discovery process ignores those adjacent intents, you'll overbuild around your internal definition of the market.
Map intent before you map features
I like to sort queries into problem-aware, solution-aware, and decision-stage buckets. Problem-aware searches reveal pain. Solution-aware searches reveal category expectations. Decision-stage searches show how people compare vendors and alternatives.
That same structure now helps with AI discovery too. Question-based searches often become prompts, and prompts often collapse categories. Someone asking for "the best tool for client onboarding" may get a very different answer than someone asking for "client portal software."
The practical workflow is straightforward:
- List core jobs: Start with the recurring outcomes buyers need.
- Group by intent: Separate educational queries from shortlist and comparison queries.
- Review snippets and AI answers: Look at how your category gets summarized in public-facing results.
- Spot language mismatch: Note where users say one thing and your site says another.
The point isn't to chase every keyword. It's to understand the demand surface around your product. For many SaaS teams, discovery breaks because product, SEO, and product marketing each use different market language.
If your team wants a tactical research layer here, AI-powered keyword research can help uncover phrasing patterns that traditional keyword buckets miss.
4. Citation Source Analysis & Content Gap Identification
When an AI assistant mentions your brand, the answer didn't come from nowhere. It was shaped by a set of sources, often product docs, review sites, help articles, comparison pages, community threads, or partner content. If the answer is weak, those sources usually are too.
This is one of the most useful newer product discovery techniques because it tells you why the market sees you the way it does. You may think you're known for workflow automation, but if assistants keep citing third-party reviews that emphasize reporting, your public narrative is split.
What citation gaps usually reveal
The most common issue isn't a total lack of content. It's weak content architecture. Teams publish plenty, but the pages don't answer buyer questions cleanly, don't explain trade-offs, or bury positioning under vague copy.
A citation audit should ask:
- Which pages get cited most often: Docs, homepage, integration pages, help center, review profiles.
- Which topics are missing: Use cases, alternatives, migration paths, implementation detail, pricing logic.
- Which sources are outdated: Old docs and stale comparison pages can anchor bad summaries for a long time.
- Which competitors own trust signals: Reviews, technical docs, partner pages, and public implementation content.
MyMentions has a practical guide on auditing brand visibility on LLMs, and it aligns well with this kind of citation-first discovery work.
A useful mental model here is simple. If AI systems repeatedly cite everyone except your own materials, your product isn't just under-discovered. It's under-explained.
5. Competitive Benchmarking & Share of Voice Analysis
A buyer asks ChatGPT for the best tools in your category, then follows up with "which one is best for a mid-market SaaS team?" If your competitor shows up three times across those prompts and you appear once with a vague description, the problem is not just visibility. It is category control.
Competitive benchmarking helps teams see how the market frames them before they burn cycles on roadmap changes or homepage rewrites. In SaaS, that framing shapes shortlist inclusion, demo quality, and how often buyers treat you as a serious option versus a backup.
The useful benchmark set is usually wider than the obvious rival list. Include direct competitors, adjacent tools, and substitutes that solve the same job in a different way. That is how teams catch the comparisons buyers make, especially in AI-influenced discovery where a prompt can collapse multiple categories into one answer.
What matters in practice:
- Prompt-level presence: Which brands appear in category, use-case, alternative, migration, and "best for" prompts.
- Narrative ownership: Whether competitors get attached to terms like enterprise-ready, fast setup, strong integrations, better reporting, or lower total cost.
- Proof sources: Which public assets support those claims, such as docs, review sites, partner pages, analyst coverage, or implementation content.
- Positioning failure points: Where your product gets summarized too broadly, confused with another category, or omitted entirely.
I use this work to answer a harder question than "who gets mentioned most?" I want to know who owns the commercial narrative. A competitor with fewer total mentions can still beat you if they dominate high-intent prompts and get cited with stronger proof.
That is where share of voice becomes more useful than a basic feature matrix. MyMentions has a solid breakdown of share of market versus share of voice if you need a practical way to connect mention share to actual category position.
One trade-off matters here. Broad monitoring gives a fuller picture, but it also creates noise. Narrow benchmarking is easier to act on, but teams often miss substitute products and emerging AI-native competitors. Start with 10 to 20 high-intent prompts, compare the top five brands that appear repeatedly, then expand only after you can explain the pattern.
For SaaS teams using MyMentions, this becomes operational instead of theoretical. Track which competitors surface across prompts, which sources support them, and where your brand loses ground by use case or persona. That gives product, marketing, and sales one shared map of the market instead of three disconnected opinions.
Benchmarking should change decisions. If it does not lead to a sharper category page, a better comparison asset, new proof content, or tighter positioning, it is just monitoring.
6. User Behavior Analytics & Sentiment Tracking
A familiar SaaS failure looks like this. AI search, organic search, or a partner mention drives a wave of qualified-looking traffic. Signups rise. Activation does not. The issue usually is not traffic volume. It is a gap between what people expected before they arrived and what the product delivers in the first session.
That gap is a discovery signal.
Behavior analytics shows where interest breaks down after the click. Sentiment tracking shows why. Used together, they help teams spot bad-fit demand, misleading positioning, and onboarding friction early enough to fix them before those problems get baked into roadmap debates.
Pair behavior with perception
I want clear answers to two questions. Do users from AI-driven discovery activate, convert, or bounce differently from other channels? Did the framing they saw before arrival match the product experience they got?
For modern SaaS teams, that second question matters more than it used to. AI assistants often summarize your product before a buyer sees your site. If those summaries overstate ease of use, integrations, or ideal use cases, users arrive with the wrong mental model. They churn faster, leave harsher feedback, and create false signals that can look like a product problem when the underlying issue is expectation mismatch.
Track this with a setup your team can maintain:
- Channel cohorts: Compare AI-referred visitors with search, partner, community, and direct traffic.
- Activation funnels: Watch where users stall in signup, onboarding, workspace setup, invite flows, or first-value actions.
- Expectation match: Review the language users saw before the click, then compare it with trial feedback, sales notes, and onboarding responses.
- Post-signup sentiment: Ask a simple question early: what did you expect this product to help you do?
MyMentions is useful here because it gives context for the pre-click side of the journey. If your brand keeps getting surfaced in prompts around one use case, but those visitors fail to activate and mention a different expectation in feedback, you have a positioning problem to fix. Sometimes that means updating onboarding. Sometimes it means changing the proof points and claims that AI systems keep picking up. Sometimes it means accepting that you are attracting the wrong segment.
That trade-off matters. Broad sentiment collection gives you more signal, but it also mixes together very different buyer types. Tight cohorting by source, persona, or use case gives cleaner insight, but sample sizes stay smaller. I usually start narrow, then widen only after I can explain the pattern with both behavioral data and actual user language.
Good teams treat sentiment as an early indicator of product-market fit by segment, promise accuracy, and message quality. That makes analytics useful for discovery, not just reporting.
7. Stakeholder Feedback & Internal Alignment
Internal feedback is messy, political, and often biased. It's still valuable. Sales hears objections before product does. Customer success hears where promises collide with reality. Support hears confusion in its rawest form. Product marketing hears category friction long before a roadmap item gets approved.
The mistake is taking each team's requests at face value. The better move is to treat internal input as directional evidence that needs validation.
Turn internal noise into usable evidence
I like structured collection here. Ask each function the same kinds of questions: what prospects misunderstand, what customers repeatedly ask for, what competitors get credit for, and where positioning falls apart.
Then synthesize patterns instead of reacting to volume. If sales says enterprise buyers want audit logs, support says admins are confused by permissions, and onboarding analytics show drop-offs in workspace setup, those signals may point to the same underlying issue.
A few patterns make this process work:
- Use a shared template: Force consistency in how teams report problems and evidence.
- Separate anecdotes from patterns: One loud deal shouldn't outrank repeated friction.
- Bring artifacts: Call recordings, support tickets, lost-deal notes, onboarding transcripts.
- Close the loop: Show teams which signals turned into research, experiments, or roadmap decisions.
The teams that get this right don't ask for alignment at the end. They build it into discovery from the start.
8. Content Strategy & Positioning Refinement
A prospect lands on your site after seeing your product in ChatGPT, Perplexity, or a peer recommendation. They understand the problem. They still cannot tell whether your tool fits their workflow, what makes it different, or why they should trust it. That is usually a positioning failure disguised as a traffic problem.
For SaaS teams, content does three jobs at once. It helps buyers evaluate fit. It gives AI systems cleaner language to cite and summarize. It forces the product team to state, in plain terms, who the product is for and where it is not the best choice.
Turn discovery signals into content decisions
Good positioning rarely starts with a blank page. It starts with evidence. Interview notes, prompt testing, citation analysis, sales call transcripts, and win-loss reviews already show where understanding breaks down. Content should close those gaps.
I use a simple scoring pass for the content backlog. Rate each topic on likely buyer impact, confidence in the signal, and effort to ship. Then prioritize the pages that remove real friction fast. A clear comparison page or implementation guide often does more for pipeline than another broad opinion post.
A few content formats consistently earn their place:
- Use-case pages: Match the product to a specific workflow, team, or trigger event.
- Comparison pages: Explain where you win, where you do not, and what trade-offs buyers should expect.
- Decision-stage docs: Answer adoption questions like setup time, permissions, integrations, security review, and migration effort.
- Problem-first education: Lead with the buyer's problem statement, not the category label your team prefers.
This matters more in AI-mediated discovery. If an assistant is trying to answer, "What tool helps a RevOps team monitor brand mentions in AI answers?" it needs explicit source material. Vague homepage copy does not help. A focused page that explains the use case, the workflow, and the proof points does.
That is where teams using MyMentions have an advantage if they build around the signals they already collect. Prompt testing shows how models describe the category. Citation analysis shows which sources shape that description. Your content strategy should respond directly to both. If AI assistants keep citing review sites for category language but miss your implementation depth, publish pages that make your differentiation easier to quote and easier to verify.
If you need an operating model for that workflow, this guide to AI content strategy for SaaS teams is a useful starting point.
Strong positioning does not try to sound bigger. It tries to sound clearer. In practice, clarity wins more often.
9. Partner & Community Ecosystem Development
Some products become easier to discover because of what they connect to, not just what they say about themselves. Slack, Zapier, Figma, Stripe, and Notion all benefit from ecosystems that create repeated mentions across docs, partner pages, templates, tutorials, and communities.
That matters in AI-mediated discovery because assistants often pull confidence from distributed signals. If your product shows up across integration guides, community recommendations, and implementation examples, it's easier for both people and models to understand where you fit.
Authority often comes from the edges
Partnerships work best when they reinforce a specific use case. A generic partner logo grid does almost nothing for discovery. A detailed integration page, a co-authored workflow guide, or a partner-owned implementation example often does much more.
For SaaS teams, I look for ecosystem opportunities in three buckets:
- Integration authority: Products your customers already trust and use daily.
- Community authority: Forums, creator communities, and practitioner spaces where workflows get discussed.
- Implementation authority: Agencies, consultants, and service partners who explain how the product is used in the wild.
This technique also helps validate market language. If partners consistently describe your product differently than you do, pay attention. They may be closer to buyer understanding than your homepage is.
Community language is often a better predictor of discoverability than internal messaging docs.
10. Conversion Optimization & Feedback Loop Implementation
A common SaaS miss looks like this. Discovery produces strong interview notes, useful prompt tests, and sharper positioning. Traffic rises, signups stay flat, activation stalls, and the team still cannot point to the insight that changed buyer behavior.
Good discovery changes the funnel, not just the slide deck. AI-era teams need a working connection between what appears in prompts, citations, search queries, and sales conversations, and what users see on the page, in the product, and during onboarding.
The operating question is simple: how fast can your team turn a signal into a change that improves conversion or activation?
Twenty Ideas makes this point well in its article on measuring product discovery success. Repeated customer contact matters. What matters more operationally is whether those findings change copy, onboarding steps, qualification logic, pricing presentation, or product defaults. If they do not, discovery turns into observation instead of progress.
A useful feedback loop has four parts:
- Source-aware entry points: Match page messaging to the channel that created the visit. Users arriving after an AI mention usually need quick proof, category clarity, and a concrete use case before they will keep reading.
- Intent capture at signup: Ask one lightweight question about the job they need done, the tool they may replace, or what triggered the search.
- Activation review: Compare the acquisition promise with first-session reality. If users arrive expecting monitoring, attribution, or AI visibility insights, the product needs to confirm that value early.
- Experiment linkage: Turn repeated objections and drop-off patterns into named hypotheses, owners, and test windows.
Teams usually fail in the handoff. They update a headline and leave onboarding untouched. They collect churn reasons and never compare them with the promises made in search snippets, AI answers, partner pages, or demo calls.
For a tool like MyMentions, the loop can be very concrete. If prompt testing shows that buyers find you through brand monitoring or AI citation tracking use cases, the homepage, signup flow, and first report should reinforce that path. If win-loss notes show confusion between mention tracking and broader social listening, treat it as a conversion problem, a qualification problem, and possibly a packaging problem.
I use the Double Diamond as an operating model here: discover, define, develop, deliver. It gives product, growth, and content teams a shared way to decide whether a signal calls for clearer messaging, a workflow change, or a shipped feature.
Speed matters. Weekly review cycles usually beat quarterly retrospectives because the signal is still fresh and the owner is still close to the problem. A tight set of connected inputs, prompt performance, page conversion, activation friction, and win-loss reasons, gives SaaS teams a much better chance of turning discovery into revenue.
Product Discovery Techniques: 10-Point Comparison
A comparison table is useful, but SaaS teams rarely fail because they picked a bad technique. They fail because they run one method in isolation, then treat the output like a decision instead of a signal.
The better approach is to use this table as a sequencing tool. Start with the method that matches your current bottleneck. If MyMentions needs better visibility in AI assistants, begin with prompt testing and citation analysis. If traffic is strong but activation is weak, behavior analytics and conversion feedback deserve priority. If the team is debating positioning, interviews, search intent work, and stakeholder alignment will usually produce better answers than another feature brainstorm.
The trade-off is speed versus confidence. Lighter methods such as stakeholder input or interview rounds can surface direction quickly, but they also carry more bias. Heavier methods such as attribution analysis, competitive monitoring, and ongoing prompt testing take more setup, yet they give teams a clearer read on whether demand, discoverability, and conversion are improving.
I use the table to make three decisions:
- Which signal should we trust first?
- Which method is expensive enough to require an owner and cadence?
- Which combination gives us evidence we can act on this quarter?
That matters more now because product discovery for SaaS no longer stops at customer research and keyword lists. Teams also need to understand how AI systems describe the category, which sources they cite, what prompts trigger visibility, and whether that visibility turns into qualified pipeline. For a product like MyMentions, that makes discovery more operational. The work connects market language, AI discoverability, product expectations, and revenue impact.
Use the comparison to choose a starting point. Then build a system, not a stack of disconnected activities.
From Discovery to Delivery Your Action Plan
The best product discovery techniques don't live in separate playbooks. They reinforce each other. Prompt testing shows how AI assistants frame your product. Interviews explain why buyers use certain language. Search intent research shows what people are trying to solve before they know your name. Citation analysis exposes why competitors get recommended. Funnel and cohort reviews show whether visibility leads to meaningful usage.
That's the system SaaS teams need now. Not a one-time research sprint, and not a backlog full of loosely connected ideas. A real discovery system combines quantitative signals, customer evidence, competitive context, and message testing in a loop that keeps running.
The "Discovery Sandwich" is still one of the clearest operating principles. Start with data. Look for the anomaly, the drop-off, the odd pattern, or the category confusion. The Product Management Society gives a simple example of a 20 to 30 percent sudden drop in engagement as a trigger for deeper research. Then move into qualitative work to understand what users are experiencing. Then return to the data to confirm whether the explanation holds up. That sequence keeps teams from overreacting to anecdotes or hiding behind dashboards.
I also like putting tight limits on ideation. Amplitude recommends brainstorming sessions of around 30 minutes, which is a useful constraint for cross-functional teams. It forces quantity, keeps people from overdefending early ideas, and helps you move toward experiments instead of endless discussion.
If you're deciding where to start, don't try all ten techniques at once. Pick one foundational method and one amplifier. Often, this means customer interviews plus either prompt testing or citation analysis. If your positioning is weak, start there. If your traffic is healthy but conversion is soft, focus on behavior analytics and onboarding feedback. If every department has a different view of the market, fix internal alignment before shipping another major initiative.
Strong discovery also requires a better standard for hypotheses. The Product Management Society notes that a strong hypothesis should define the problem, propose a solution, and predict a measurable outcome within a triangulated discovery process. That's a good bar. If your team can't articulate those three pieces, it probably isn't ready to build.
The companies that win in SaaS don't just build fast. They learn fast, update their understanding fast, and tighten the loop between market signals and shipped work. In 2026, that loop has to include AI-mediated discovery. Buyers are asking assistants to shortlist products, compare alternatives, and explain categories. Your discovery practice needs to reflect that reality.
Start small. Run a prompt set. Talk to customers. Audit the citations shaping your brand. Check where users drop off. Then turn those findings into a backlog your team can ship against.
If your team wants a practical way to monitor AI-driven discovery, MyMentions helps you track how assistants discover, rank, cite, and describe your product across major AI platforms. It turns prompt-level visibility, sentiment, and citation data into concrete fixes your product, SEO, and marketing teams can act on.