Storm AI Form Fill is designed to answer questions from new inputs — voice notes, photos, and typed text. A related question is whether Storm can read fields that are already filled in on the same form and consolidate them into a separate summary field, without any new input being provided.
Direct Answer
Yes — Storm can attempt this, though results are most reliable on smaller forms and with manual verification of the output.
Storm can consolidate existing fields into a summary field, but may produce errors in some sections even when the form is built consistently. Two failure types have been observed: silent merge (content attributed to the wrong field) and silent drop (content missing entirely, with no warning). Errors are more likely on larger forms.
At a Glance
Storm can consolidate existing same-form fields into a summary field, but may produce errors in some sections even when the form structure is identical across all sections.
Two failure types observed: silent merge (content from one item written into the wrong field) and silent drop (content missing from the output entirely, with no blank or warning).
These failures look like correct output — unlike Storm's usual behaviour of leaving a field blank when unsure.
Accuracy gets worse on larger forms. The more fields Storm has to process at once, the more likely it is to mix up or miss content.
Works best when the assessor manually verifies the output against source fields before the record is finalised — especially for compliance, QMS, or audit records.
Reliability improvements are planned after GA as part of Storm v2. The Applied AI team may be able to improve results short-term through better prompting.
What Storm Is Designed to Do
Storm's form fill is built around processing new inputs. It generates questions from the form's structure and answers them using audio, photos, video, and text you provide. The two patterns it is designed and documented for are:
Progressive re-fill on the same form — running Storm again on a form it has already partially filled. Storm reads its previous output and adds to or amends it based on new input. Existing content is preserved unless you explicitly instruct Storm to change it.
Cross-form referencing — using a previously completed form in the same folder as context (for example, filling a timesheet from daily diaries in the same folder). Storm notes which form informed each field.
Storm reads existing field content as context to inform other fields, but is not designed to copy content verbatim from one field into another. Consolidating already-completed fields across multiple sections into a separate summary field works best on smaller forms — results become less reliable as form size and field count increase.
What Works
When asked to consolidate existing fields on the same form, Storm can:
Group output by section — when your prompt explicitly instructs Storm to group by section (for example, "group by section 1.1, 1.2, 1.3, 1.4"), Storm follows that structure correctly.
Skip blank sections — sections where nothing was entered are correctly left out of the output rather than shown as empty placeholders.
Reproduce field content accurately in some sections — when Storm maps fields correctly, it reproduces the content as written, including correctly pairing each Required Action to its source Finding or Concern.
Known Failure Modes
Storm's standard behaviour when it is unsure about content is to leave the field blank for human review. When consolidating existing same-form fields, two additional failure modes have been observed that are different from the standard blank-if-unsure behaviour:
Silent merge — text from one item (for example, a Concern's Required Action) is written onto the end of a different item's field (for example, a Finding's Required Action), with no separator. The merged item's own field then appears missing. The output looks complete but the content is incorrectly placed.
Silent drop — content that is present in the source field does not appear anywhere in the output. It is not left blank and not merged elsewhere — it is simply absent, with no error or placeholder shown.
Both failure modes produce output that appears complete but contains errors. This is different from Storm leaving a field blank, which signals that review is needed. Silent merge and silent drop do not signal anything — the output looks finished.
These failures have been observed to occur inconsistently across sections with identical field structures. The same structure in one section may produce correct output while failing in another, and this cannot be predicted or designed around at the template level.
Why This Happens
Storm processes all fields in a single pass across the entire form. When a form has many sections with the same structure — for example, Finding → Required Action, Concern → Required Action repeated across four or more sections — Storm has to keep track of which content belongs to which field while holding the whole form at once. In some cases, Storm produces output it is confident enough to fill in, but with the content placed in the wrong field or missing entirely.
The longer and more complex the form, the more likely these errors are to occur. Storm v2 plans to break the process into smaller steps to reduce this.
Prompting Guidance That Helps
The following prompt patterns improve output quality for this use case:
Explicit section grouping — instruct Storm to group by section name or number (for example, "group by section 1.1, 1.2, 1.3, 1.4") rather than by field type. Without this, Storm may restructure the output by field type across sections instead of section by section.
Verbatim instruction — if exact reproduction is required, state it explicitly: "reproduce verbatim — do not paraphrase."
Blank-skipping instruction — tell Storm to omit sections or fields where nothing was entered. This works correctly and keeps the output clean.
Note: Even with clearer prompting, merge and drop failures can still occur. These happen because of how Storm processes many similar fields in a single pass, not because of unclear instructions. The Applied AI team may be able to improve prompt-level handling for this pattern short-term.
Does Accuracy Hold at Scale?
Accuracy is expected to get worse as the form gets larger.
Testing on a small form (4 sections, identical field structure) produced errors in 2 of 4 sections. The Storm team has confirmed that the same pattern on a significantly larger form will produce more errors, not fewer.
As field count grows, Storm has to process more content at once, which creates more opportunity for the same merge and drop errors to occur. Test on the actual form you plan to use before relying on this pattern as part of a customer build — a small-form test may not reflect how Storm performs on a larger, full-size form.
When to Use This Pattern — and When Not To
Works well when | Take extra care when |
The form has a small number of sections | Using the output as a compliance, QMS, or audit record without checking it against source fields |
The assessor will manually verify the output against source fields before the record is finalised | A missing or incorrectly placed Required Action would affect the accuracy of the compliance record |
The summary is low-stakes and occasional errors can be caught and corrected in review | The form is large — accuracy gets worse as field count and form size increase |
If You Encounter Errors
Re-run Storm on the same form. The failure is inconsistent — a re-run may produce a correct result for the affected section.
Consolidate one section at a time. Running Storm on one section per session reduces the number of similar fields it has to process at once and may improve accuracy.
Verify against source fields. After any Storm consolidation run on compliance-critical content, check each item in the output against its source field before treating the record as final.
Log feedback in #storm-feedback. The Storm team is tracking how often this pattern comes up. Submit via #storm-feedback with the session trace.
Related articles: