Custom GPTs/models
Brandwatch Boolean Query Builder
Descriptions
Build boolean queries based on topics
Instructions
See /Users/azntaiji/Seafile/Developer/research/etc/gpt-boolean-builder.md
MG Patient Conversations Analyst
Description
Analyze and summarize patient conversations about MG drugs and therapies
Instructions
# Section 1: Overall Role
Analyze and summarize provided texts (usually Facebook Groups posts and comments) to help me better understand patient sentiment and experiences surrounding drugs and therapies indicated for MG (Myasthenia Gravis), including but not limited to: VYVGART Hytrulo (subcutaneous), VYVGART (intravenous), SCIG and SCIG brands (HIZENTRA, HYQVIA, PANZYGA), SOLIRIS (eculizumab), RITUXAN (rituximab), ULTOMIRIS (ravulizumab), RYSTIGGO (rozanolixizumab), ZILBRYSQ (zilucoplan), IVIG and IVIG brands (GAMMAGARD, GAMMAPLEX, GAMUNEX-C, PRIVIGEN), and IMAAVY (nipocalimab).
# Section 2: Sentiment Analysis Instructions
- Note that multiple drugs and therapies may be mentioned with varying experiences in each post/comment, for example: "I was on rituxan for 2 years and that is all I needed. Now on vyvgart hytrulo; it's slowing down progression but not stopping it. I'll likely have to go back on rituxan." and "I just had my third vyvgart hytrulo. It is working and I’m so relieved. I do feel some fatigue. But compared to the hizentra twice a week it’s so much better.".
- For these types of post and comments, overall sentiment should be analyzed based on the primary drug/therapy being discussed. And if you cannot determine the primary drug/therapy being discussed, base sentiment on the patient's experience with VYVGART or VYVGART Hytrulo, if either is mentioned.
# Section 3: General Rules
- Use bullet points for any lists
- Bold important terms
- Preface any headlines or titles with relavant emojis for extra flair
- Format headlines and titles in AP title case style
- Format descriptions in regular sentence capitalizaton/case
- Headlines, titles, and descriptions should be quantified whenever possible
- When quote, post, or comment examples are requested: Never alter, alter, truncate, or modify the original texts
# Section 4: How to Think and Respond to Prompts
When asked to analyze and generate key themes:
1. Create 3-4 key theme headlines (≤60 characters each)
2. Each theme should include a description (≤150 characters each)
3. Include 4-5 related verbatim quotes under each description, bulleted
When asked to identify shifts in conversations over time, i.e. month-over-month or quarter-over-quarter:
1. Create 3-4 key theme headlines (≤55 characters each)
2. Within each key theme, include the following, using the below format:
- "`TIME PERIOD 1` Focus:" with 100-character description. Below this, include 4 related verbatim posts/comments as bullets.
- "`TIME PERIOD 2` Shift:" with 150-character description. Quantify/include shift data in the description. Below this, include 4 related verbatim posts/comments as bullets.
When asked to generate presentation-ready headlines for provided texts:
1. Combine the provided texts into 6–8 concise headlines (≤180 characters each) that summarize all the provided texts
2. Each headline should integrate every theme into one cohesive narrative, rather than giving separate headlines per theme. Keep them clear, analytical, and presentation-readyConversation Starters
- Identify key themes discussed about
DRUG_THERAPYwithin the attached texts. - Identify month-over-month shifts in conversations regarding
DRUG_THERAPYwithin the attached texts. - Identify quarter-over-quarter shifts in conversations regarding
DRUG_THERAPYwithin the attached texts. - Generate presentation-ready headlines for the below texts:
CIDP Patient Conversations Analyst
Description
Analyze and summarize patient conversations about CIDP drugs and therapies
Instructions
# Section 1: Overall Role
Analyze and summarize provided texts (usually Facebook Groups posts and comments) to help me better understand patient sentiment and experiences surrounding drugs and therapies indicated for CIDP (Chronic Inflammatory Demyelinating Polyneuropathy), including but not limited to: VYVGART Hytrulo (subcutaneous), SCIG and SCIG brands (HIZENTRA, HYQVIA, PANZYGA), RITUXAN (rituximab), IVIG and IVIG brands (GAMMAGARD, GAMMAPLEX, GAMUNEX-C, PRIVIGEN), and IMAAVY (nipocalimab).
# Section 2: Sentiment Analysis Instructions
- Note that multiple drugs and therapies may be mentioned with varying experiences in each post/comment, for example: "I was on rituxan for 2 years and that is all I needed. Now on vyvgart hytrulo; it's slowing down progression but not stopping it. I'll likely have to go back on rituxan." and "I just had my third vyvgart hytrulo. It is working and I’m so relieved. I do feel some fatigue. But compared to the hizentra twice a week it’s so much better.".
- For these types of post and comments, overall sentiment should be analyzed based on the primary drug/therapy being discussed. And if you cannot determine the primary drug/therapy being discussed, base sentiment on the patient's experience with VYVGART or VYVGART Hytrulo, if either is mentioned.
# Section 3: General Rules
- Use bullet points for any lists
- Bold important terms
- Preface any headlines or titles with relavant emojis for extra flair
- Format headlines and titles in AP title case style
- Format descriptions in regular sentence capitalizaton/case
- Headlines, titles, and descriptions should be quantified whenever possible
- When quote, post, or comment examples are requested: Never alter, alter, truncate, or modify the original texts
# Section 4: How to Think and Respond to Prompts
When asked to analyze and generate key themes:
1. Create 3-4 key theme headlines (≤60 characters each)
2. Each theme should include a description (≤150 characters each)
3. Include 4-5 related verbatim quotes under each description, bulleted
When asked to identify shifts in conversations over time, i.e. month-over-month or quarter-over-quarter:
1. Create 3-4 key theme headlines (≤55 characters each)
2. Within each key theme, include the following, using the below format:
- "`TIME PERIOD 1` Focus:" with 100-character description. Below this, include 4 related verbatim posts/comments as bullets.
- "`TIME PERIOD 2` Shift:" with 150-character description. Quantify/include shift data in the description. Below this, include 4 related verbatim posts/comments as bullets.
When asked to generate presentation-ready headlines for provided texts:
1. Combine the provided texts into 6–8 concise headlines (≤180 characters each) that summarize all the provided texts
2. Each headline should integrate every theme into one cohesive narrative, rather than giving separate headlines per theme. Keep them clear, analytical, and presentation-readyConversation Starters
- Identify key themes discussed about
DRUG_THERAPYwithin the attached texts. - Identify month-over-month shifts in conversations regarding
DRUG_THERAPYwithin the attached texts. - Identify quarter-over-quarter shifts in conversations regarding
DRUG_THERAPYwithin the attached texts. - Generate presentation-ready headlines for the below texts:
[OLD] MG and CIDP Patient Conversations Analyst
Description
Analyze and summarize patient conversations about MG and CIDP drugs and therapies
Instructions
Analyze and summarize provided texts (Facebook Groups posts and comments) to help me better understand patient sentiment and experiences surrounding drugs and therapies indicated for MG (Myasthenia Gravis) and CIDP (Chronic Inflammatory Demyelinating Polyneuropathy), including but not limited to: VYVGART Hytrulo, VYVGART, SCIG and SCIG brands (HIZENTRA, HYQVIA, PANZYGA), SOLIRIS (eculizumab), RITUXAN (rituximab), ULTOMIRIS (ravulizumab), RYSTIGGO (rozanolixizumab), ZILBRYSQ (zilucoplan), IVIG and IVIG brands (GAMMAGARD, GAMMAPLEX, GAMUNEX-C, PRIVIGEN), and IMAAVY (nipocalimab).
Note that multiple drugs and therapies may be mentioned with varying experiences in each post/comment, for example: “I was on rituxan for 2 years and that is all I needed. Now on vyvgart hytrulo; it’s slowing down progression but not stopping it. I’ll likely have to go back on rituxan.” and “I just had my third vyvgart hytrulo. It is working and I’m so relieved. I do feel some fatigue. But compared to the hizentra twice a week it’s so much better.”. For these types of post and comments, overall sentiment should be analyzed based on the primary drug/therapy being discussed. And if you cannot determine the primary drug/therapy being discussed, base sentiment on the patient’s experience with VYVGART or VYVGART Hytrulo, if either is mentioned.
Generally, you should provide analyses and summaries in a bulleted outline format, and include 2-5 verbatim quotes as sub-bullets for context. Feel free to add emojis for extra flair.
Conversation Starters
- Identify key themes discussed within these texts
- Summarize these texts
Prompts
Argenx
To outline key themes for VYVGART and Hytrulo Facebook Groups slides:
Use prompt with excel file of mentions that has column headers
Date,Thread Title,Post/Comment, andSentiment. Make sure to include current and previous month's data, and sort the columns by Post/Comment then Thread Title to group posts.
Analyze the attached texts for mentions about VYVGART or Hytrulo and identify a shift in conversations from `MONTH1` to `MONTH2`, outlined as 4 key themes. `OPTIONAL1: Try and highlight `SENTIMENT` shifts in sentiment specifically regarding VYVGART or Hytrulo.` `OPTIONAL2: Try and highlight discussions about `TOPICS`, but only if there are any.` Each key theme should use the following format:
- 45-55 character theme title in AP style
- "`MONTH1` Focus:" with 100-character description. Below this, include `NUMBER` verbatim quotes as bullets. (do not alter or truncate the original texts for the quotes)
- "`MONTH2` Shift:" with 100-character description. Below this, include `NUMBER` verbatim quotes as bullets. (do not alter or truncate the original texts for the quotes)v2 for quarterlies: Keep only Quarter, Post/Comment, Sentiment Columns—make sure to change Date column header to Quarter and use Q$ 202$ values instead of actual dates.
Analyze the attached posts/comments ("Post/Comment" column) for discussions about `DRUG` and identify a shift in conversations from `TIMEPERIOD1` to `TIMEPERIOD2` ("Quarter" column), outlined as 4 key themes. 2 of the themes should highlight `SENTIMENT` shifts in sentiment specifically regarding `DRUG`. The other 2 themes should highlight general conversation shifts and/or positive discussions regarding `DRUG`.
Each key theme should use the following format:
- 45-55 character theme title in AP style
- "`TIMEPERIOD1` Focus:" with 100-character description. Below this, include 4 related verbatim posts/comments as bullets. (do NOT alter or truncate the posts/comments)
- "`TIMEPERIOD2` Shift:" with 100-character description. Below this, include 4 related verbatim posts/comments as bullets. (do NOT alter or truncate the posts/comments)Helpful prompt for discussions of competitor drugs and therapies:
Use prompt with excel file of mentions that has column headers
Date,Thread Title,Post/Comment, andSentiment. Sort the columns by Post/Comment then Thread Title to group posts.
Standard prompt:
Analyze the attached texts for mentions specifically about TREATMENTS and generate 3 key themes. `Optional: Highlight conversations that are SENTIMENT regarding TREATMENTS.` Each key theme should use the following format:
- 115-130 character theme title, not in title case, but in standard sentence structure and regular font weight. Make use of bold only to highlight parts of the title that stand out.
- `NUMBER` verbatim quotes as bullets (do not alter or truncate the original texts for the quotes)Prompt to include separate themes for multiple therapies:
Analyze the attached texts for mentions specifically about TREATMENTS and generate 2 key themes for each. `Optional: Highlight conversations that are SENTIMENT regarding TREATMENTS.` Each key theme should use the following format:
- 115-130 character theme title, not in title case, but in standard sentence structure and regular font weight. Make use of bold only to highlight parts of the title that stand out.
- `NUMBER` verbatim quotes as bullets (do not alter or truncate the original texts for the quotes)NEW!
New format
Prompt for new shifts format (starting with Q3 2025):
Analyze the attached posts/comments ("Post/Comment" column) for discussions about `DRUG_THERAPY` and identify a shift in conversations from `TIMEPERIOD1` to `TIMEPERIOD2` ("`DATETYPE`" column), outlined as a key theme.
The key theme should use the following format:
- 45-55 character theme title in AP style
- "`TIMEPERIOD1` Focus:" with 125-character description.
- "`TIMEPERIOD2` Shift:" with 150-character description. Quantify/include shift data in the description. Below this, include 4 related verbatim posts/comments as bullets. (do NOT alter or truncate the posts/comments)LEO Pharma
For ANZUPGO websites analytics recommendations slide:
Generate 5 website recommendations based on the below data. Each recommendation should include a title under 40 characters and a description under 180 characters. Recommendations should focus on a goal of optimizing the website and improving on-site engagement:
`PASTETAKEAWAYSFROMOTHERSLIDESHERE`Prompt for CHE/Hand Eczema monthly reporting platforms slides title and subtitles:
For ChatGPT:
Summarize the below text into a headline and subtitle for a slide about discussions taking place on `platform`. Headline should be 180 characters or less, in AP style title case. Subtitle should be 90 characters or less, using normal sentence capitalization. Provide 3 different options for headline/subtitle combinations:
`Add all text under headline and subtitle here`For Gemini (in Slides, while on the slide:
Summarize the text currently on this slide into a headline and subtitle for a slide about discussions taking place on `platform`. Headline should be 180 characters or less, in AP style title case. Subtitle should be 90 characters or less, using normal sentence capitalization. Provide 3 different options for headline/subtitle combinations.
Prompt for CHE/Hand Eczema Facebook Groups slide to generate key themes and quotes on the right side:
Identify 4 discussion themes within the below texts, which are based on patient experiences with different treatments for hand eczema. Then outline each discussion theme in the following format: Title in 40 characters or less (in AP style title case), description in 130 characters or less (normal sentence case), followed by 1 verbatim quote example.
Texts:
`Paste in all mentions here`LEO: For website overview slide key takeaways
Summarize `MONTH` trends for `WEBSITE`, including a month-over-month comparison.
Metrics:
- Don't include conversions as a metric
- Use Session primary channel group (sessionPrimaryChannelGroup) as the primary dimension for measuring session data
- When querying total sessions, total users, new users, etc., aggregate without a dimension (aggregate at the date level) to avoid duplication
Include 4-5 key takeaways and 1-2 recommendations. Below are 3 examples from previous months—please use a similar formatting:
# August 2025 Key Takeaways
- User growth is up 3% MoM — paid social remains the top traffic driver (35% of all sessions). 64% of paid social traffic comes from Facebook, 36% comes from LinkedIn.
- Users coming from paid social and organic search are spending more time on the site — +18 seconds and ~+2 minutes.
- Paid search represents 31% of all sessions (+5 percentage points MoM) — among
Google ads search keywords with 200+
sessions, those showing the strongest
MoM growth include "topical steroid treatment" (+224%), "long term atopic dermatitis" (+24%) and "adbry" (+12%).
- The HCP site remains the top referrer (96% of all referral traffic).
- Recommend: Review landing page performance and conversion funnels to ensure engagement translates into leads
# July 2025 Key Takeaways
- User growth is down nearly 9% MoM. Paid social is the top traffic driver, accounting for 36.5% of all sessions. 57% of paid social traffic comes from Facebook, 42% comes from LinkedIn.
- Referral traffic is up nearly 16% (11.5% of all sessions)—with the Adbry DTC website driving 93.5% of that traffic.
- Paid search (Google Ads) traffic represents 25.8% of of all sessions. Among Google ads search keywords with 200+ sessions, those showing the strongest MoM growth include "adbry" (+15.4%) and "long term atopic dermatitis" (+8.1%).
- Organic search and direct traffic grew by 12% and 11%, respectively.
# June 2025 Key Takeaways
- User growth is down 61.5%, driven primarily by a reduction of paid traffic (social, search, and display). However, user growth is on par with most other organic-focused months, including March and April.
- The average session duration reached nearly 2 minutes, increasing 118%—driven primarily by paid search traffic, display and banner ads, and referral traffic.
- Paid social remains the #1 source of traffic (38.9% of all sessions). 85% of paid social traffic comes from Facebook, 15% comes from LinkedIn.
- Referral traffic increased 2.2% and now accounts for 9.6% of all sessions. The primary source is the DTC website adbry.com, contributing 95.7% of all referral traffic. Other minor referrers—each representing less than 1%—include zhireimbursementportal.com, app.tonohealth.com, chatgpt.com, and email.hcplive.com.
-Paid search (Google Ads) represents 27% of all traffic. Among Google ads search keywords with 300+ sessions, those showing the strongest MoM growth include “long term atopic dermatitis treatment” (+20.8%) and “ruxolitinib cream” (+20.1%).LEO: For Content Engagement slide key takeaways
Summarize `MONTH` trends for `WEBSITE` by the pagePath dimension, including a month-over-month comparison.
Metrics:
- Don't include conversions as a metric
- Include Sessions, averageSessionDuration, and EngagementRate as metrics
- Also include metrics for the following key events: enrollment_form_english_download, enrollment_form_english_download, and aggregate these key events without a dimension (aggregate at the date level) to avoid duplication
Include 2-3 key takeaways and 1-2 recommendations. Below are 3 examples from previous months—please use a similar formatting:
# August 2025 Key Takeaways
- The Dosage and Administration page drives the most traffic (33%) — overtaking home page traffic in August. Users are also spending more time on this page, over one minute.
- Users spend the most time on the Safety Results page — increasing 5.3x to nearly 11 minutes.
# July 2025 Key Takeaways
- The homepage continues to receive themost traffic (30% of all traffic), but traffic to the Adult Dosage and Administration Page increased by 32.8x.
- Users spend the most time on the Support and Resources page (overn ine minutes), while average times pent on the page grew by 181%.
# June 2025 Key Takeaways
- The Adult Efficacy Results page remains the leading page by traffic (37.9% of all traffic), but the Pediatric Safety and Pediatric Efficacy Results pages saw the strongest growth—up 30.8% and 24.1%, respectively.
- Users spent the most time on the Adult Dosage and Administration page.Abbott
Generating a stacked bar chart:
Row 1 is the column header. Group the Company column into 3 segments:
- Abbott: Based explicity on `Abbott` Company row
- Others: Based explicity on `Uncategorised` Company row
- Competitors: Based on all other Company rows (excluding Abbott and Uncategorised)
Then use the grouped data, and create a stacked bar chart with columns B:I on the X axis, and the Company column on the Y axis.
- Sort the Y axis by SUM of values in descending order
- Sort the X axis by value in descending order
- Use the following hex codes to color the Company bars:
- Abbott: #009CDE
- Others: #D9D9D6
- Competitors: #004F71
output python code:
# Load and process the Excel data, then create the requested stacked bar chart.
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from pathlib import Path
from caas_jupyter_tools import display_dataframe_to_user
# --- Load ---
xlsx_path = Path("/mnt/data/data.xlsx")
if not xlsx_path.exists():
raise FileNotFoundError("Expected /mnt/data/data.xlsx but it was not found.")
df = pd.read_excel(xlsx_path, header=0)
# --- Identify columns ---
# Try to find the "Company" column case-insensitively. Fallback to the first column if not found.
company_col = None
for c in df.columns:
if str(c).strip().lower() == "company":
company_col = c
break
if company_col is None:
company_col = df.columns[0]
# Columns B:I are positional indices 1..8 (inclusive)
if df.shape[1] < 9:
# If the sheet doesn't have columns through I, fall back to using whatever numeric columns exist after the company column,
# up to 8 columns total, to avoid crashing; still follow the intent.
value_cols = [col for col in df.columns if col != company_col]
# Prefer numeric columns
numeric_cols = [c for c in value_cols if pd.api.types.is_numeric_dtype(df[c])]
# Cap to at most 8 columns
value_cols = numeric_cols[:8] if numeric_cols else value_cols[:8]
else:
value_cols = list(df.columns[1:9])
# Ensure numeric
for c in value_cols:
df[c] = pd.to_numeric(df[c], errors="coerce")
# --- Grouping logic ---
def classify_company(name: str) -> str:
n = str(name).strip().lower()
if n == "abbott":
return "Abbott"
if n == "uncategorised":
return "Others"
return "Competitors"
df["_Group"] = df[company_col].map(classify_company)
# Aggregate sums for each group across B:I
grouped = df.groupby("_Group")[value_cols].sum(numeric_only=True).reindex(["Abbott", "Others", "Competitors"])
# Ensure all groups exist (fill missing with zeros)
for g in ["Abbott", "Others", "Competitors"]:
if g not in grouped.index:
grouped.loc[g] = 0
grouped = grouped.loc[["Abbott", "Others", "Competitors"]] # enforce order
# --- Sort X (columns) by total value across all groups, descending ---
col_totals = grouped.sum(axis=0)
sorted_cols = list(col_totals.sort_values(ascending=False).index)
grouped_sorted_cols = grouped[sorted_cols]
# --- Determine group order by total sum (to partially satisfy "sort Y axis by SUM desc") ---
group_totals = grouped_sorted_cols.sum(axis=1).sort_values(ascending=False)
group_order = list(group_totals.index)
grouped_final = grouped_sorted_cols.loc[group_order]
# --- Plot: stacked bars where X = columns (B:I), stacks = Company groups with specified colors ---
colors = {
"Abbott": "#009CDE",
"Others": "#D9D9D6",
"Competitors": "#004F71",
}
fig, ax = plt.subplots(figsize=(12, 6))
bottom = np.zeros(len(sorted_cols))
x = np.arange(len(sorted_cols))
# Plot groups in order of total descending so larger segments are drawn first
for g in group_order:
vals = grouped_final.loc[g].values.astype(float)
ax.bar(x, vals, bottom=bottom, label=g, color=colors.get(g))
bottom += vals
ax.set_xlabel("Columns (B:I)")
ax.set_ylabel("Value")
ax.set_title("Stacked Bar: Columns B:I on X, Company groups as stacks")
ax.set_xticks(x)
ax.set_xticklabels(sorted_cols, rotation=45, ha="right")
ax.legend(title="Company Group")
ax.margins(x=0.01)
plt.tight_layout()
# Save figure
out_png = "/mnt/data/stacked_bar.png"
fig.savefig(out_png, dpi=150, bbox_inches="tight")
# Also provide a tidy table of the grouped data in the exact plot order
tidy = grouped_final.copy()
tidy.loc["TOTAL"] = tidy.sum(axis=0)
display_dataframe_to_user("Grouped data (Companies × Columns B:I, with totals)", tidy)
out_png
General
Prompt to generate headlines/slide titles (Copy and paste all texts to summarize after the prompt):
New version 2025-10-23 refined via ChatGPT:
Combine the following texts into 6–8 concise, AP title case headlines (≤180 characters each) that summarize all the provided text. Each headline should integrate every theme into one cohesive narrative, rather than giving separate headlines per theme. Keep them clear, analytical, and presentation-ready.Optional add-on 1: for insight decks or market research:
Make the tone data-driven, using metrics or directional phrasing like ‘rises,’ ‘falls,’ or ‘shifts.’Optional add-on 2: for executive or brand presentations:
Make the tone strategic and forward-looking, emphasizing market momentum and key behavioral shifts.OLD VERSION:
Generate 8 possible headlines in 180 characters or less each, for the below texts, in AP title case style. This will be for a slide about `TOPICHERE`:Follow-up prompts on generated themes:
Anything else interesting from Theme `TITLE`?Do a deeper dive into Theme `TITLE`For theme `TITLE`, what are the most frequently discussed infusion cycles and durations?