diff --git a/subjects/ai/nlp-scraper/README.md b/subjects/ai/nlp-scraper/README.md
index 3be89edd9..9b210ce8f 100644
--- a/subjects/ai/nlp-scraper/README.md
+++ b/subjects/ai/nlp-scraper/README.md
@@ -143,7 +143,6 @@ project
 ├── nlp_enriched_news.py
 ├── README.md
 ├── results
-│   ├── topic_classifier.pkl
 │   ├── enhanced_news.csv
 │   └── learning_curves.png
 └── scraper_news.py
@@ -169,7 +168,8 @@ python scraper_news.py
 2. Run on these 300 articles the NLP engine. The script `nlp_eneriched_news.py`
    should:
 
-   - Save a `DataFrame` with the following struct:
+   - Save a `DataFrame` with the following struct and store the result in a
+     `csv` file, `enhancend_news.csv`:
 
    ```
    Unique ID (`uuid` or `int`)
@@ -215,10 +215,6 @@ python scraper_news.py
    Environmental scandal detected for <entity>
    ```
 
-> I strongly suggest creating a data structure (dictionary for example) to save
-> all the intermediate result. Then, a boolean argument `cache` fetched the
-> intermediate results when they are already computed.
-
 ### Notions
 
 - [Web Scraping](https://www.youtube.com/watch?v=XVv6mJpFOb0)
diff --git a/subjects/ai/nlp-scraper/audit/README.md b/subjects/ai/nlp-scraper/audit/README.md
index 7609587b7..6dd031407 100644
--- a/subjects/ai/nlp-scraper/audit/README.md
+++ b/subjects/ai/nlp-scraper/audit/README.md
@@ -18,11 +18,15 @@
 
 ###### Are the learning curves provided?
 
-###### Do the learning curves prove the topics classifier is trained correctly - without overfitting?
+###### Do the learning curves prove the topics classifier is trained correctly - without overfitting? Ask the student to explain what the term "overfitting" means and how he avoided this phenomenon.
+
+> Additionally, you can look for external resources. For example, Wikipedia has a good page on "overfitting".
+
+##### Ask the student to train and store the topic classifier model in a file named `topic_classifier.pkl`.
 
 ###### Can you run the topic classifier model on the test set without any error?
 
-###### Does the topic classifier score an accuracy higher than 95%?
+###### Does the topic classifier score an accuracy higher than 95% on the given datasets?
 
 ##### Scandal detection