Getting Your Data Out of Qualtrics: Exports, Formats, and What Breaks

TL;DR#

A Qualtrics data export moves your raw response rows out cleanly and leaves most of your derived work behind. Closed-ended answers, open-text verbatims, and response metadata come out as CSV, TSV, XLSX, XML, SPSS, or — via the Response Import/Export API — JSON and NDJSON. Survey structure exports only as QSF, a proprietary Qualtrics JSON format that contains no responses, no workflows across brands, and no translations. Display logic, branch conditions, Text iQ topic models, dashboard definitions, calculated metrics, and role permissions have no portable export at all — they get re-authored by hand in the new platform. Hard ceilings matter: exports should stay under 1.8 GB, a dataset cannot exceed 100,000 columns, dashboard widget exports cap at 10,000 rows in CSV/TSV/XLSX and 500 rows in PDF, and Text iQ allows a maximum of 1,000 topics per question. Run the full export 60 to 90 days before renewal, not after you give notice — once the contract lapses, the interface you need to run the export is the thing you just cancelled.

Start the Qualtrics Data Export Before You Start the Negotiation#

The export is leverage, and it only works as leverage if you already have the files. Teams routinely reverse the order: they open the renewal conversation, get a discount pitch, decide to leave anyway, give notice, and only then discover that eleven dashboards, four years of coded verbatims, and a contact directory with transaction history all need to come out through a UI they have three weeks of access to.

Two things go wrong when the export happens after the decision. First, you lose the ability to walk. A buyer who cannot produce their own data cannot credibly threaten to leave, and vendors price accordingly — the same dynamic that shows up in Qualtrics pricing in 2026 and in the services and overage line items covered in Qualtrics implementation costs. Second, you lose the calendar. A large brand's full export is not an afternoon; it is a multi-week sequence with polling, chunking, and reconciliation.

There is also a compliance argument for doing it early. Under Article 20 of the GDPR, a data subject has the right to receive personal data "in a structured, commonly used and machine-readable format" and to transmit it to another controller without hindrance. If your customer feedback contains personal data — and if you collect email, IP address, or recipient fields, it does — the ability to satisfy a portability or erasure request cannot depend on an active vendor login. That obligation survives the contract, which is exactly the argument made in the guide to customer feedback data retention and privacy.

If you have not yet decided whether to leave, the honest diagnostic is in the signs it's time to leave Qualtrics and the cost math in is Qualtrics worth it in 2026. Export anyway. A current archive is good hygiene whether you renew or not.

What a Qualtrics Data Export Gives You Cleanly#

Qualtrics exports raw response data reliably, in open formats, with usable metadata. This is the part that works, and it is worth naming precisely so you know what you do not need to worry about.

Through Data & Analysis, you can export responses as CSV, TSV, Excel (XLSX), XML, SPSS (.sav), or straight to Google Drive. The Response Import/Export API adds JSON and NDJSON, which matters if you are loading into a warehouse rather than a spreadsheet. Every one of those except SPSS and XLSX sits on the National Archives' preferred list for structured data — NARA's transfer guidance tables name CSV, JSON, XML 1.1, ODS, ASCII text, and SIARD as preferred, with Excel formats only "acceptable." Export to CSV or JSON for the archive copy and let XLSX be the convenience copy.

The response file itself is well-formed. Every row is a response; every column is a field. CSV and TSV files ship with three header rows: the internal Qualtrics field ID (ResponseId, EndDate, Q1), the human-readable field name or question text, and an ImportId row. All three are required if you ever want to re-upload the data, so do not let a well-meaning analyst delete rows two and three to make the file "clean."

The metadata columns are genuinely useful and often overlooked:

• ResponseId — the stable join key for everything else you export.
• StartDate, EndDate, RecordedDate — start, submit, and platform-record timestamps.
• Duration (in seconds) and Progress — time on survey and percent complete.
• Finished — whether the respondent hit an endpoint.
• Status — flags previews, tests, imported, and offline responses so you can exclude them.
• DistributionChannel and UserLanguage — how they arrived and in what language.
• RecipientFirstName, RecipientLastName, RecipientEmail, ExternalDataReference — contact-list identity, when a panel was used.
• LocationLatitude, LocationLongitude — IP-derived or GPS coordinates.
• Q_LastSeenFlowID and last-seen question IDs — the final flow element and questions displayed, which is the closest thing you get to a breadcrumb of the path taken.

Export options give you real control. You choose values (numeric recodes) or labels (full answer text), whether to download all fields or only displayed ones, whether to split multi-value fields into separate columns, whether to use internal IDs in the header, whether to recode seen-but-unanswered questions as -99 and unanswered multi-value fields as 0, whether to exclude response edits, whether to include viewing-order data for randomized surveys, and whether to include download links for user-uploaded files. TSV handles special characters and non-Latin alphabets more reliably than CSV, which is the right default for any multi-market program.

Run the export twice — once with values, once with labels. Analysts want the recodes; anyone reading the archive in three years wants the words.

What Does Not Survive a Qualtrics Data Export#

Roughly speaking, anything Qualtrics computed for you does not export as something another platform can read. Raw answers are portable. Structure, logic, models, and configuration are not.

Two structural ceilings will find you before you notice them. Exports should be kept under 1.8 GB, and a dataset cannot exceed 100,000 columns. Worse, if a survey has more than 6,500 basic fields or 2,000 larger fields, the export interface restricts column selection to contact fields and survey metadata only — which means the very largest, most-instrumented surveys in your account are the hardest ones to get out. Chunk those by date range and pull them through the API instead of the UI.

Survey Logic and Branching Metadata#

Survey logic exports as a proprietary file that only Qualtrics can read, which makes it documentation rather than data. The QSF contains all your survey formatting and settings, question text, answer choices, and even questions sitting in the trash — but it is a Qualtrics-shaped JSON, and the platform's own guidance is not to edit it, because editing corrupts it.

That leaves you with three imperfect options for preserving logic, and you should do all three:

1. Export the QSF for every active survey. It is the only complete record, even if it is only readable back inside Qualtrics.
2. Export each survey to Word with logic and recode values included. This produces a human-readable specification you can hand to whoever rebuilds the instrument. It cannot be re-imported, which is fine — the audience is a person.
3. Screenshot or diagram the Survey Flow itself. Branch conditions, randomizers, quotas, and authenticator blocks are easier to re-implement from a picture of the flow than from either of the above.

The response-level consequence is subtler. When branching means different respondents saw different questions, blank cells in the export carry two very different meanings — "this question was never shown to you" and "this question was shown and you skipped it." That distinction disappears unless you turn on the seen-but-unanswered recode, which writes -99 for skipped scale items and 0 for skipped multi-value options. Turn it on. Reconstructing intent from ambiguous blanks a year later is not possible, and the difference materially changes every completion and drop-off metric you will report to leadership — the same numbers that end up on the CX scorecard for the board.

Coded Themes and Text Analytics Output#

Coded themes export as their current output, never as the model that produced them. This is the single most expensive loss in a CX platform migration, because coding is the part your team actually spent years on.

Text iQ lets you export topic lists per question, and with work you can join topic assignments back to ResponseId and keep the labels. What you cannot take is the topic model: the rules, the trained classifiers, the sentiment calibration, and the accumulated hand-corrections. Note also the ceiling — a maximum of 1,000 topics per question, and imports that would push a question past that limit are rejected wholesale rather than partially applied. And the two text products do not interoperate on the way out: an XM Discover topic export is a different file format from Text iQ's, and it cannot be imported into Text iQ. If your organization runs both, you are exporting two incompatible taxonomies.

The practical move is to stop treating the taxonomy as an asset to migrate and start treating it as a hypothesis to re-test. Export the verbatims plus the current topic labels, then re-code the corpus in the new platform and compare. That is how verbatim analysis across 40,000 open-ended responses actually gets rebuilt, and it is usually faster than trying to replicate legacy rules. If text analytics is the reason you are on the platform at all, the replacement landscape is mapped in Qualtrics Text iQ alternatives and Qualtrics XM Discover alternatives.

Dashboards, Panels, and Contact History#

Dashboards export their contents, not themselves. You can export any widget's underlying data to CSV, TSV, or XLSX up to 10,000 rows, or to PDF up to 500 rows — but the dashboard definition, its filters, its calculated metrics, its date logic, and its role-based visibility rules produce no configuration file. Every executive view gets rebuilt.

Budget for that explicitly. A brand with fifteen dashboards is looking at a rebuild measured in weeks, and that labor belongs in the CX platform total cost of ownership model rather than being discovered in month two. Before you export, take a full-page screenshot of every dashboard and write down what each widget computes — filter set, metric definition, and comparison period. That document is what makes the rebuild a translation exercise instead of an archaeology project.

The directory is the other quiet trap. An XM Directory contact export can include contact statistics — invite count, email count, average response time, response rate, last invite date, last response date, and response count — plus whichever embedded data fields you select. Two cautions apply. Selecting all embedded data can inflate the file dramatically in a directory with many fields, so choose deliberately. And the touchpoint timeline displays up to 100 transactions per contact, which means deep interaction history is not something you should assume you can reconstruct from the UI. In XM Directory Lite, embedded data fields are capped at 1,024 characters, so long free-text attributes may already be truncated at rest.

What no export gives you is a clean per-message send log with opens, bounces, and unsubscribes joinable to responses. Ask for it in writing before you cancel, and if it is not available, treat suppression state as something you rebuild from your ESP or CRM rather than from the CX platform. Teams migrating between the two big suites hit the identical wall — the same gap appears in the guide to switching off Medallia.

The 5-Step Qualtrics Export Sequence, With Timing#

Run the export as five phases over roughly 45 days, starting 90 days before renewal. The timing is the point: each phase surfaces problems that the next phase needs time to solve.

Step 1 — Inventory (Days 1–3). Count everything: active and archived projects, surveys, dashboards, directories, mailing lists, workflows, and API integrations. Flag any survey with more than 6,500 basic fields or 2,000 larger fields, because those will not export normally and need an API path. Record the response count per survey per month — that number is your reconciliation target later. Pull the same inventory into the 60-day CX platform migration checklist so the export plan and the migration plan share one source of truth.

Step 2 — Export the structure (Days 4–10). QSF plus Word-with-logic for every survey you might ever rebuild. Screenshots of every Survey Flow and every dashboard. A written data dictionary mapping question ID to question text to answer options to recode values. NARA's guidance is blunt on this point: structured data transfers need accompanying schemas and data dictionaries to be verifiable. Your future self needs the same thing.

Step 3 — Export the responses (Days 11–20). Two passes per survey — one with values, one with labels. Prefer TSV for any multi-language program. Turn on split multi-value fields, seen-but-unanswered recoding, viewing-order data for randomized blocks, and download links for uploaded files. Keep each file under 1.8 GB by chunking on date range. For anything large, drive the Response Import/Export API instead: start the export job, poll the progress endpoint until percentComplete hits 100, then fetch the compressed file. Poll on an interval rather than in a tight loop — brands have a concurrent API request ceiling, and brand administrators can set tighter per-user limits on top of it.

Step 4 — Export the derived layer (Days 21–30). Text iQ topic lists per question. Dashboard widget exports in 10,000-row chunks. Directory contacts with statistics and a deliberate embedded-data selection. Any actual uploaded files, fetched through the download links while the account is live. This is the phase that always takes longer than planned, which is why it sits before validation rather than after.

Step 5 — Validate and archive (Days 31–45). Reconcile, checksum, store in open formats, and only then walk into the renewal meeting. If you are running a parallel evaluation, this is also the window where running a CX platform pilot and scoring vendors against a weighted CX vendor scorecard fit — you will make better shortlist decisions holding your own data than describing it from memory.

How to Validate the Export Before You Cancel#

Validation means proving the export matches the source, not confirming that a file downloaded. Work this list before notice goes out, while you still have the account to re-run anything that fails.

• Row-count parity. Compare exported row counts to platform response counts per survey, per month. Account for previews, tests, and imported responses using the Status column before you call a gap a gap.
• Column parity. Every question in the QSF should have at least one column in the response file. Matrix questions spread across many columns and loop-and-merge repeats the question set per iteration — expect the column count to exceed the question count, and confirm the mapping rather than the total.
• Verbatim integrity. This is where silent corruption lives. CSV, as RFC 4180 states plainly, "has never been formally documented," and there are "considerable differences among implementations." Fields with embedded line breaks, commas, or double quotes must be quoted and internal quotes doubled — and a parser that gets it wrong shifts every subsequent column without raising an error. Parse the file programmatically, assert the row count matches, and eyeball the ten longest verbatims.
• Date sanity. Confirm the brand time zone and normalize timestamps to an unambiguous YYYY-MM-DD ordering per ISO 8601 before loading anywhere. Locale-dependent date parsing is the second most common silent failure after quote handling.
• Attachment fetch. Actually download ten uploaded files and open them. Do not accept a column of links as evidence that the files are yours.
• Freeze test. Re-run one survey's export a week later and diff it. If the numbers move, you have late-arriving or edited responses and need a documented cutoff date.
• Written deletion terms. Get the vendor's deletion commitment in writing, including backups. Deleted data's disaster-recovery backups are removed within 90 days, and free accounts default to a 6-month retention window — but your contract, not the help center, governs. Align the language with recognized practice such as NIST Special Publication 800-88 Revision 2 on media sanitization, finalized in September 2025, and keep the confirmation with the archive.

What to Do With the Data Once It's Out#

An export is an archive, not a research program — the listening has to restart somewhere. The files answer questions about what customers said through 2026. They do not tell you what customers think next quarter, and every day between cancellation and a working replacement is a day of feedback you never collected.

This is the moment to fix the underlying problem rather than port it. Most of what broke in the export broke because the platform stored answers to fixed fields and kept the intelligence — the coding, the logic, the dashboard math — in proprietary configuration. Perspective AI takes the opposite shape: AI-led interviews where customers answer in their own words, the follow-up questions are generated in the moment instead of pre-wired into branch logic, and the analysis layer reads transcripts rather than depending on a hand-tuned topic taxonomy. There is no branching graph to lose because there is no branching graph.

Practically, that means a much smaller migration surface. Start by replacing your highest-volume relationship survey with a conversation — the voice of customer survey template and the customer journey interview template both map directly onto programs that currently run as CoreXM instruments. For churn and renewal signals, the churn interview template covers the ground a transactional NPS study usually misses. Perspective is built for CX teams running this exact transition, and the wider category shift — from survey suite to conversational research — is the argument in what a customer experience platform is and why AI is replacing the survey suite and in how enterprise feedback management became what it is.

If you are still shortlisting, write your requirements before you look at demos — the CX platform requirements checklist and the RFP questions worth asking vendors are the right order of operations, and the eight Qualtrics alternatives worth evaluating in 2026 is the shortest path to a candidate list. Module-specific replacements are broken out for CoreXM, CustomerXM, and EmployeeXM.

Frequently Asked Questions#

How do I export all my data from Qualtrics?#

Export in four separate passes, because no single action captures everything. Pull response data from each project's Data & Analysis tab or the Response Import/Export API; pull survey structure as a QSF plus a Word document per survey; pull dashboard contents widget by widget; and pull contacts from the directory. Users, roles, workflows, dashboard definitions, and text-analytics models have no export and must be documented manually.

What file formats does Qualtrics export?#

Qualtrics exports response data as CSV, TSV, Excel (XLSX), XML, SPSS (.sav), or directly to Google Drive, and the Response Import/Export API adds JSON and NDJSON. Survey structure exports only as QSF, a proprietary format, or as a non-reimportable Word document. For an archive you expect to read in five years, prefer CSV or JSON — both appear on the National Archives' preferred-format list for structured data.

Does a Qualtrics data export include survey logic and branching?#

No portable export of survey logic exists. Display logic, branch conditions, randomizers, and quotas are encoded inside the proprietary QSF file, which only Qualtrics can read, and the Word export renders logic as human-readable text that cannot be re-imported anywhere. Plan to re-author every condition by hand, and enable the seen-but-unanswered recode so branch-driven blanks in your response file stay distinguishable from genuine skips.

Can I export Text iQ topics and sentiment from Qualtrics?#

You can export the topic lists and their current assignments, but not the model that generated them. Topic rules, trained classifiers, sentiment calibration, and years of manual corrections do not transfer. Text iQ caps each question at 1,000 topics, and an XM Discover topic export uses a different file format that cannot be imported into Text iQ — so brands running both products are exporting two incompatible taxonomies.

How long does Qualtrics keep my data after I cancel?#

Retention after cancellation is governed by your contract, not by a published default. Backups of data you delete are removed within 90 days, and free accounts carry a default 6-month retention window, but paid brands vary. The vendor's stated position is that customers are responsible for their own backups and retention policies, so treat the export as your obligation and get the post-termination deletion timeline in writing.

Should I export values or labels from Qualtrics?#

Export both, in separate files. Values give you the numeric recodes analysts need for statistical work and warehouse loading; labels give you the full answer text that anyone reading the archive later can actually interpret. Because the choice is made per export rather than per column, running two passes is the only way to preserve both, and it costs one extra job per survey.

Next Steps#

A Qualtrics data export is two jobs, not one: getting the rows out, which the platform does well, and reconstructing everything the platform computed on top of them, which it cannot help with at all. Raw responses, verbatims, and metadata come out in open, well-documented formats. Survey logic, coded themes, dashboards, calculated metrics, panel history, and permissions do not — they are documentation projects and rebuild work, and they are the reason a realistic export sequence runs 45 days rather than an afternoon.

Do it in this order: inventory, structure, responses, derived layer, validate. Start 90 days before renewal so the export is finished before the negotiation starts, and validate with row-count parity, verbatim parsing, and a real attachment fetch before notice goes out. Then pair the export plan with the full Qualtrics migration playbook and get the buying committee aligned on what is being rebuilt versus retired.

Once the archive is safe, the question stops being "how do we get our data out" and becomes "what should we be asking customers now." That is worth answering with conversations instead of another form. Start an AI-led customer interview with your own questions and see what a transcript captures that a dropdown never did — or review Perspective AI pricing to model the swap against your current contract.