ChatGPT–Canva integration: repeated failures, missing previews and blocked website development
I have been using ChatGPT with Canva to develop a website and branded content for my business. Across multiple chats, normal sessions and Work Mode sessions, I have experienced repeated problems that have delayed the work and left important parts unfinished.
Some problems appear to be limitations of the integration. Others involve ChatGPT’s execution and reporting. From a user’s perspective, both need to be addressed because they affect whether we can complete and maintain the work.
1. Approval is requested before the design can be viewed
ChatGPT repeatedly refers to “the preview above” or says that pages have been “visually presented”, but no preview appears for me on Windows, iOS or Android. It then asks me to approve or save changes I cannot inspect.
In one website correction, the missing preview created a deadlock: I could not approve the unseen footer, and ChatGPT reported that it could not retain the unsaved draft while waiting. It cancelled the changes. A later workaround involved saving to a separate review copy before I could inspect the result in Canva.
Requested fix: Provide reliable previews before requesting approval, with a working fallback review link. Preserve drafts across messages. Clearly distinguish saving a review draft, approving a design and publishing it.
2. Work is created in a format that does not meet the editing requirement
I needed an editable Canva website. Earlier versions were described as Canva Code designs that exposed no ordinary editable text or image elements to the connection. Conversion attempts reportedly disrupted spacing and layout, leading to a proposed manual rebuild.
A footer was also replaced with a flattened composite image over the native elements. I only discovered this when I tried to edit it.
Requested fix: Confirm the output format and its editing limitations before construction. Preserve text, buttons, navigation, links and layout components as native editable elements. Do not silently substitute flattened images for editable sections.
3. Essential website construction and functionality remain unsupported
During development, ChatGPT reported that the available connection could not reliably create required page layouts, global sections, product-card structures or internal links. Other unresolved items included technical element locks, product carousels and purchasing interactions.
The Cocktail Edit link returned “Page not found”, and the Explore the Edits button remained unresolved. Contact-form email delivery was also unfinished.
Requested fix: Check the actual capabilities available for the specific design before starting. Clearly identify unsupported requirements and provide a dependable execution route instead of discovering blockers after substantial work.
4. Existing video content is reported as empty
For a Canva Reel containing content, ChatGPT repeatedly reported that the connector returned is_empty: true with no editable timeline elements. This prevented direct corrections such as consistent logo positioning between scenes and pushed work back to manual editing.
Requested fix: Reliably expose existing video scenes and timeline elements. If the format is unsupported, report that accurately rather than treating the design as empty.
5. Unnecessary copies create clutter and uncertainty
Eight accidental duplicate website designs were created during one Home-page rebuild. Some had titles matching legitimate reference designs, so identifying them safely required exact design IDs.
The connection could not move them to Trash. They were eventually archived, but I was warned they could still appear in Recent views.
Requested fix: Update the designated working design unless duplication is necessary and explained. Clearly identify the current master and provide supported cleanup of accidental copies.
6. Approved decisions are not consistently preserved
I repeatedly had to restate approved colours, typography, navigation, imagery restrictions and page architecture. Generated brand designs failed to follow explicit instructions, and previously rejected treatments influenced further work.
Construction instructions also referenced an outdated Canva design until I corrected the source. Layout problems—including overlapping text, off-centre button labels and cut-off blocks—continued to require my intervention.
Requested fix: Maintain the current source of truth across chats and execution modes. Apply targeted changes without disturbing approved elements, and perform a complete visual and functional check before presenting work for review.
7. Planning outputs are substituted for requested implementation
After I authorised a Canva website build, ChatGPT initially delivered a Word build pack and uploaded assets without making the requested live page changes. I had to explain again that I wanted construction in Canva.
I did separately request some planning documents, and those were appropriate. The problem is substituting a document when the authorised task is implementation.
Requested fix: Keep the requested deliverable explicit. If construction is blocked, explain that before diverting the work into another format.
8. Work Mode handoffs and completion reporting are unreliable
Moving between normal chats and usage-limited Work Mode sessions added handoff work and uncertainty. One execution chat became exhausted while important functions remained unfinished.
A later save instruction authorised transferring changes, but the handover could not confirm that the transfer had completed. It was difficult to distinguish what existed in a draft, what was saved in a review copy and what had reached the master.
Requested fix: Preserve execution state across sessions and provide a verified record of the exact design edited, changes saved, outstanding items and next step. Instructions to save must not be reported as proof that saving succeeded.
These issues have caused repeated correction cycles, rebuilding, cleanup and interruptions. We have had to stop development runs, and the website remains incomplete.
I would like the integration to support a dependable workflow where I can specify the work, inspect it, approve it and continue from the same editable master. At present, too much time goes into supervising the process and recovering from errors instead of completing the website.