Blog

Synthetic respondents vs surveys vs real consumer opinions

Use a survey when you must measure how many people hold a view. Use public opinions to learn what people already said and why, and treat synthetic answers as a draft that you check with people.

What is the difference between surveys, synthetic respondents and public opinions?

A survey asks people. A synthetic respondent imitates people. Public opinions record what people chose to say on their own, without being asked.

  • Survey responses are answers people give to questions you wrote. Most come from a panel, which is a paid pool of survey takers. The answers are prompted: people respond because you asked.
  • Synthetic respondents are answers that an AI model writes as if it were a person in your target group. No person answers. The model predicts a typical answer from the text it learned from.
  • Public opinions are what people say about products without being asked. They appear in videos and what is said in them, posts, comments, reviews, forums and articles. Each one has a person and a source behind it.

The three sources also differ in cost and speed. A survey costs money for each completed answer and takes days or weeks to field. Synthetic answers arrive in minutes at a low cost. Public opinions already exist, so the work is in finding, reading and grouping them.

Which source should you use for which question?

Match the source to the type of question. The table shows where each source is strong.

Question you need answeredBest sourceWhy
How many people hold a view?SurveyYou control who answers, so you can count them.
Will people buy, and at what price?SurveyStated intent and price need a controlled question.
How do people react to an idea that is not on sale?Survey or interviewsNobody can post about a product they have never seen.
Why do people choose a product already on shelf?Public opinionsPeople explain their reasons in their own words.
What do people complain about or wish for?Public opinionsComplaints and requests appear unprompted, with detail.
Which words do buyers use?Public opinionsYou read the exact phrases people use.
Which of many early ideas deserve a test?Synthetic respondents, then peopleA fast screen is useful if you confirm the shortlist.

Two rules follow from the table. Public opinions find needs and reasons, but they do not size them. Surveys size things well, but they work best when you already know what to ask.

That is why the sources work better together than alone. Public opinions tell you what to ask. Surveys tell you how many people agree.

Are synthetic respondents accurate enough for consumer research?

They are good enough for some early tasks, but not for decisions. A model predicts the typical answer from the text it learned from. That has three effects.

  • It smooths out disagreement. Real groups split in untidy ways. Synthetic answers tend to cluster around the middle, so minority views can disappear.
  • It has no lived experience. A model cannot tell you how your product felt in someone’s hands last week. It can tell you what people usually write about products like it.
  • It is weakest where little is written. A new product, a niche category or a small audience leaves little text to learn from.

One published test points the same way. In a study reported in July 2026, the insight group Strat7 compared answers from synthetic data companies with answers from real people. It found the synthetic answers significantly different on many measures.

These effects matter most for the questions insights teams care about. A team wants to know where its buyers differ from the average and what went wrong last month. Those are the answers a model is least able to give.

Use synthetic respondents to draft a questionnaire, test the wording of questions or screen a long list of ideas. Then check the result with people before you commit budget. We treat a synthetic answer as a hypothesis, not a finding.

Can public posts and reviews answer the same questions as a survey?

Some of them, but not all. Public opinions can answer part of a questionnaire before you field it, so the survey you pay for is shorter.

They answer well when the question is about a product people already use and discuss. Why did you choose it? What disappoints you? What did you compare it with? What do you use it for? People answer these questions in public every day, in detail.

They answer badly when the question needs a count of a population. Public opinions have no fixed sample. You do not know how many people chose to stay silent. Heavy posters can dominate a topic, and people with strong views post more than people with mild ones.

Three habits keep the answers honest:

  1. Count people, not posts. One person who posts many times is still one person.
  2. Write the base. Say “X of the N people who say they used it”, never a share of all buyers.
  3. Label what you read. Mark each post as organic or sponsored, then decide what to filter.

What should you do about survey panel fraud?

Treat it as one error among several, not as a reason to drop surveys. Panels face fake respondents, bots, careless answers and AI-written open answers. Studies measure these problems in different ways, so no single rate describes them.

Public opinions carry their own errors. Some reviews are fake. Some posts are paid for. Groups of accounts can push one view. Every source has errors, so the useful question is which errors each source has.

Use the sources to check each other:

  • Write screeners from public language. Questions in people’s own words are harder to fake and easier to answer.
  • Compare open answers with public opinions. If panel answers sound nothing like what product users say in public, look closer.
  • Check surprising survey results. If a result conflicts with what people say unprompted, find out why before you act.
  • Keep the sources apart in your notes. Say which answer came from a survey, which from public opinions and which from a model.

How do you combine surveys, synthetic answers and public opinions?

Run them in sequence, starting with the lowest cost. Each step narrows what the next step must do.

  1. Write the decision. One sentence: what will you do differently, depending on the answer?
  2. Turn it into a goal or hypothesis. For example: “Lapsed buyers leave because the refill is hard to find.”
  3. List the questions. Write every question a survey would ask about the goal.
  4. Check public opinions for each question. Grade each one: strong evidence, some evidence, thin evidence or needs a survey. Our preliminary research guide shows how.
  5. Write answers for the strong questions. Give the answer, the spread of views, linked quotes and the caveats.
  6. Use synthetic answers where they fit. Draft wording or screen ideas, and mark every result as unconfirmed.
  7. Field a shorter survey. Keep the questions that need a count, a price or a reaction to a new idea.
  8. Go deeper with people. Interview vetted product users where the reasons are still unclear.
  9. Compare the sources. Where the survey and public opinions disagree, find out why before you decide.

The grade in step 4 is the most useful output. A question graded “needs a survey” is a good result, because it shows where to spend your budget.

Take the refill example. Public opinions might show strong evidence that buyers complain about finding refills in stores. They might show some evidence that buyers switch to a rival with a subscription. A synthetic screen could help word the answer options. The survey then measures how many lapsed buyers share each reason, and interviews explore the ones nobody expected.

Where does Honestly fit?

Honestly is a proprietary dataset of product opinions, built from many data streams. It includes videos and what is said in them, posts, comments, replies, reviews and forums. Most sources are public. The record is ours: each opinion keeps its source link. Most opinions are tied to a product, and many to a specific feature.

For consumer insights teams, this is preliminary research. Bring a goal or a hypothesis. No questionnaire is needed. Before any survey, we look for the questions people already answer in public. Next, you field a shorter survey for the rest. Then go deeper with people who used the product.

Honestly labels the posts it detects as sponsored, and your team decides what to filter. To see how we publish studies, read our research.

Frequently asked questions

Can synthetic respondents replace real survey respondents?

No. They can help draft questions and screen ideas, but no person stands behind each answer. Check any result you will act on with real people.

Can public opinions answer a whole survey?

Usually not. They answer questions about products people already use and discuss, such as reasons, complaints and comparisons. Questions about counts, prices and new ideas still need a survey.

When should I use synthetic respondents?

Use them for early, low-risk tasks: drafting questions, testing wording and screening many ideas. Confirm anything you will act on with people.

Is survey panel fraud a reason to stop running surveys?

No. Every source has errors, including public opinions. Use the sources to check each other, and write screeners in the words people use in public.

How do I give numbers from public opinions?

Count distinct people, not posts, and always give the base. Write “X of N people who say they used it”, never a share of all buyers.

Can I use synthetic answers and public opinions together?

Yes, if you keep them apart in your notes. Public opinions give you real reasons and words, and a synthetic screen can help you draft and test wording around them.

Bring one product question. See what the dataset already holds on it.

Request a demo