You wired up an AI chatbot in FlutterFlow. The API Call > Response & Test tab shows a perfectly good answer. But on screen, the chat bubble is empty — or the action silently completes and nothing happens. This is the single most common FlutterFlow chatbot bug, and it almost always traces to one of three causes: a wrong JSON path, an error body you never see, or a malformed auth header. Let's fix each one.
Why the test panel lies to you
The Response & Test panel shows the raw HTTP body. Your UI shows whatever the JSON Path extracts from that body. When those two disagree, you get a blank reply even though the request technically succeeded. So before touching anything else, confirm what the OpenAI Chat Completions endpoint actually returns:
{
"id": "chatcmpl-abc123",
"object": "chat.completion",
"model": "gpt-4o-mini",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "Hi! How can I help you today?"
},
"finish_reason": "stop"
}
],
"usage": { "prompt_tokens": 19, "completion_tokens": 9, "total_tokens": 28 }
}
The reply text lives at choices → array index 0 → message → content.
Cause 1: The JSON path is wrong
This is the flutterflow chatbot blank response classic. In the API Call's JSON Path field, FlutterFlow uses JSONPath syntax. The correct path is:
$.choices[0].message.content
The paths that quietly return null:
$.choices.message.content— forgetschoicesis an array. You must index it with[0].$.choices[0].content— skips themessageobject.$.message.content— there is no top-levelmessage.
If your path returns null in the Test sub-tab too (not just the app), the path is the culprit. Set the variable name (e.g. reply), data type String, and confirm the preview shows the answer text.
The stringified-JSON trap
If you asked the model for JSON output (via response_format), then content is itself a string containing JSON, not a nested object. FlutterFlow hands you the raw string — you cannot drill into it with another JSON path like $.choices[0].message.content.answer. Pull content as a String first, then parse it in a custom action with jsonDecode. Trying to path-extract into a stringified field is the #1 reason flutterflow json path chatbot reply searches exist.
Cause 2: The error body is hidden, so blank looks like success
When OpenAI returns a 400 or 429, the body isn't a choices array — it's an error object:
{
"error": {
"message": "You didn't provide an API key. You need to provide your API key in an Authorization header using Bearer auth.",
"type": "invalid_request_error",
"code": null
}
}
Your JSON path $.choices[0].message.content finds nothing in that shape, so the variable is null and the bubble renders empty. The action looks like it worked. To stop guessing, replace the built-in API call with a custom action that inspects the status code and returns the real error text. Add http: ^1.2.0 under Custom Code > Settings > Pub Dependencies, then:
import 'dart:convert';
import 'package:http/http.dart' as http;
// Custom Action: getWidgetChatReply
// Returns the assistant reply, OR a readable error string — never blank.
Future<String> getWidgetChatReply(
String userMessage,
String apiKey,
) async {
final uri = Uri.parse('https://api.openai.com/v1/chat/completions');
late http.Response response;
try {
response = await http.post(
uri,
headers: {
'Content-Type': 'application/json',
// .trim() prevents a stray space/newline from corrupting the header.
'Authorization': 'Bearer ${apiKey.trim()}',
},
body: jsonEncode({
'model': 'gpt-4o-mini',
'messages': [
{
'role': 'system',
'content': 'You are WidgetChat, a friendly in-app support assistant.'
},
{'role': 'user', 'content': userMessage},
],
}),
);
} catch (e) {
return 'Network error: $e'; // timeout, no connection, bad URL
}
// Surface the hidden error instead of returning empty.
if (response.statusCode != 200) {
try {
final err = jsonDecode(response.body);
final msg = err['error']?['message'] ?? response.body;
return 'API error ${response.statusCode}: $msg';
} catch (_) {
return 'API error ${response.statusCode}: ${response.body}';
}
}
final data = jsonDecode(response.body) as Map<String, dynamic>;
final choices = data['choices'] as List<dynamic>?;
if (choices == null || choices.isEmpty) {
return 'No choices returned. Raw body: ${response.body}';
}
final content = choices[0]['message']?['content'] as String?;
if (content == null || content.trim().isEmpty) {
final reason = choices[0]['finish_reason'];
return 'Blank reply (finish_reason: $reason). '
'A content filter or token limit may have truncated it.';
}
return content.trim();
}
Now bind the action's return value straight to your chat bubble's text. Whatever happens, the user (and you) see something: the reply, the API's own error message, or a finish-reason hint. No more flutterflow api call empty mystery.
Cause 3: 400 'no API key' even though you pasted the key
The flutterflow chatgpt api 400 no api key error means OpenAI received an Authorization header it couldn't parse. In a built-in API Call, the header value must be exactly:
Authorization: Bearer [apiToken]
The traps that produce a 400 or 401:
- Variable-name mismatch. If you typed
[apiToken]in the header but named the variableapi_key, FlutterFlow sends the literal textBearer [apiToken]. The brackets must match a defined variable exactly. - An empty value. If the variable resolves to blank, OpenAI receives
Bearer(trailing space, no key) — which is precisely the "you didn't provide an API key" 400. - Including
Bearerin the value too. If your variable already containsBearer sk-...and the header template also prependsBearer, you sendBearer Bearer sk-.... - A pasted newline or space. Copying a key from a doc can append
\n. The.trim()in the custom action above defends against this; in the built-in call, re-paste carefully.
In the custom action, pass the key as a String argument and store it in App State (or better, a backend proxy) — never hard-code production keys into client builds.
A 60-second debugging checklist
- Test tab shows text but UI is blank? → JSON path. Use
$.choices[0].message.content. - Test tab returns null too? → path syntax or a stringified-JSON content field.
- Status code isn't 200? → swap in the custom action and read the error body.
- 400 'no API key'? → check the variable name match and that the key isn't empty.
finish_reasonislength? → raisemax_tokens. Ifcontent_filter, rephrase the prompt.
Or skip the plumbing
A blank reply is one of many bugs a hand-built AI chat has to survive. JSON paths, hidden error bodies and auth headers are only the start: conversation history and keeping the provider key out of the app come next.
Widget-Chat is a drop-in AI chat widget for Flutter and
FlutterFlow. Add the widget_chat package,
pass your project key, and the whole chat screen is there: answers from your own content, chat history with pagination, and a model provider key that never ships in the app.
ChatWidget(
configuration: BotConfiguration(
projectSecretKey: 'your-project-secret-key',
userID: 'user-123',
),
)
In FlutterFlow the package goes in as a custom widget. The steps are in the FlutterFlow guide.
Create a free Widget-Chat account and have it running in your app today.




