AI-moderated interviews vs surveys: which to use when

AI-moderated interviews can’t fully replace an employee engagement survey, and they shouldn’t try to. A survey gives you the same numbers, asked the same way, year after year, so you can see a trend. An AI interview gives you the reasons behind those numbers, in the words your people use. Most organisations get the best results by keeping a short survey for the trend and running interviews when they need to know why.

Can AI interviews replace employee engagement surveys?

Only if you’re prepared to give up what surveys are good at. An engagement survey is a measuring instrument. Every person gets the same items on the same scale, so a score of 3.8 in the finance team this spring can be set against 4.1 last spring and against the warehouse today. That comparability is the whole point.

An interview works differently. The questions adapt to the person. If a nurse says handovers have got worse, the interviewer asks what changed and when. The next nurse might talk about rotas instead. You learn more from each person, but two answers are never quite the same measurement.

So the honest answer is that the two methods do different jobs. If your board expects an engagement score every year, keep the survey. If your engagement score dropped and nobody can explain why, a survey won’t tell you, and an interview will.

What does an engagement survey do well?

It tracks change over time, which is the big one. Because the items stay fixed, you can tell whether a score moved after a reorganisation, a new manager or a pay round.

It uses tested scales. Well-built engagement items have been checked for whether they measure what they claim to. Gallup, for example, says it has studied over 3.3 million workers across 100,000+ teams in building its Q12 survey, and offers comparison against industry benchmarks.

It is quick for the person answering. Twenty items on a phone take a few minutes. That matters in a depot or on a ward where nobody has a spare half hour.

What a survey can’t do is follow up. A low score on “I have the tools I need to do my job” tells you there’s a problem. It doesn’t tell you whether the problem is a broken scanner, a slow approval process or a system nobody was trained on. The comment box helps a little, but most comments are short, and many people leave it empty.

What do AI-moderated interviews add that a survey can’t?

An AI-moderated interview is a one-on-one conversation run by software. The person answers an open question, and the interviewer asks a follow-up based on what they just said. It runs with everyone you invite, at the same time, so you get interview depth across a whole department instead of from a sample of ten.

The main gain is reasons. Instead of “communication 3.2”, you hear that shift leads find out about changes from the drivers, because the briefing email goes out after the early shift has left. That is something a manager can act on.

You also get people’s own words. People describe problems in their own terms, and those terms often show you how they see the organisation. When three teams independently call a new system “the black hole”, that tells you something a rating scale never will.

And interviews catch what you didn’t think to ask. A survey can only cover the topics you chose in advance. In an open conversation, people raise the thing you didn’t know about. For more on why people often say more to an AI interviewer, see the research behind AI-moderated interviews.

Spradley works this way. People get an invitation by email or Slack, answer on their own phone or computer, by text or voice. The interviewer asks a follow-up question on every answer, and the report groups what people said into themes with the quotes behind them, so leaders can compare departments.

What happens after the interviews?

In Spradley, the report lands within days: themes with the quotes and reasons behind them, compared across departments, and followed across interview rounds over time. It exports to PowerPoint, PDF, Word or Google Docs.

Then each leader gets a Personal Agent. Ask anything in plain language and every claim cites the report and the quote it rests on. A claim check removes any figure or quote that doesn’t match the evidence, and the agent says plainly what the evidence does not show. It drafts the next step: a readout, a team conversation guide that quotes no one, or a deck. When the evidence is missing, it drafts a follow-up interview on one to three themes, which the leader sends after one confirmation, and the new report joins the others. When a leader saves what they committed to, reminders come at four and eight weeks until it is done. The agent never sees one person’s answer, never shows numbers or ratings for fewer than five people, cuts sentences that name an employee, and refuses to identify anyone, to pick a team to cut or to monitor loyalty or intent to leave.

What should you plan for with AI interviews?

They are real, and you should plan for them.

It is still self-report. An interview captures what people are willing and able to tell you. It doesn’t observe the work. If someone believes the bottleneck is in purchasing when it is really in their own team, the interview records the belief.

People can choose not to answer. Participation should be voluntary, and some people will skip a question or give a short reply. That is their right. It also means the people who answer may not be a perfect mirror of the people who don’t, and a survey has the same problem.

Interviews follow themes, not a single score. You follow each theme across rounds: how often it comes up, how people describe it, and the reasons behind it.

The analysis has to show its evidence. A summary that says “morale is low in logistics” is only useful if you can see the quotes it came from. If a tool gives you conclusions without the underlying answers, you can’t check whether it read people correctly. Ask any vendor to show you how each theme links back to what was said.

Confidentiality needs care. Open answers are more identifying than ticks on a scale. A good tool hides results for small groups and strips names from quotes. We cover that in detail in how the anonymity threshold works.

When should you use a survey and when an interview?

Use a survey when you need a number you can compare: the annual engagement score, a short pulse to see whether a change is landing, or a validated scale you have to report externally.

Use interviews when the question starts with “why” or “what”. Why did the score drop? What is getting in the way after go-live? What do people in the new acquisition actually worry about? Interviews also suit small or specific groups, such as night shift or a single site, where a survey gives you too few answers to read anything into.

Use interviews before you write a survey, too. A round of open conversations tells you which topics matter, so the survey you build afterwards asks about the right things.

How do you combine surveys and interviews?

The simplest pattern is survey first, interviews second. Run your engagement survey as usual. Pick the two or three results that worry you most, and run interviews with the affected groups on exactly those topics. You keep the trend line and get the explanation within weeks instead of waiting for next year’s comments.

A second pattern is interviews between surveys. If you run an annual survey, use interviews in the gaps to check specific changes: a new shift pattern, an office move, an AI rollout. Our post on alternatives to the annual engagement survey sets out the other options.

A third pattern is triangulation. When the survey says one thing and the interviews say another, that gap is worth looking at. A team can score well on “my manager cares about me” and still describe, in interviews, a manager who is never there on the late shift. Both can be true.

Whatever the mix, close the loop. People stop answering surveys and interviews alike if they never hear what came of it. A short note back saying what you heard and what you will change does more for next year’s response rate than any reminder email. The CIPD’s employee voice factsheet notes that a quarter of employees report high levels of silence, which is a good reason to make it easy and safe to speak.

If you run HR or people analytics and want to see how interviews fit into an existing listening programme, our HR and people page shows how Spradley is used alongside surveys.

Frequently asked questions

Key terms

AI-moderated interview

A one-on-one conversation run by software, in which the interviewer asks open questions and a follow-up question based on each answer. Everyone invited can take part at the same time.

engagement survey

A survey that asks every employee the same fixed items on a scale, usually once or twice a year, so scores can be compared over time and between groups.

pulse survey

A short survey run often, typically five to ten questions on one or two themes, monthly or quarterly. It shows whether something is moving between larger surveys.

validated scale

A set of survey items that has been tested for whether it measures what it claims to, so scores can be tracked and compared.

Personal Agent

The AI agent in Spradley that works for each leader on top of what their people said. It answers questions with the report and quote behind every claim, drafts readouts, guides and decks, drafts follow-up interviews and tracks commitments, without ever seeing one person’s answer.

self-report

Information people give about themselves, in a survey or an interview. It records what people believe and are willing to say, not direct observation of the work.

triangulation

Comparing what different methods, such as a survey and interviews, say about the same question, and looking closely where they disagree.