Reply Mining for Persona Development

Collect reply and quote-tweet patterns around your brand on X, analyze behavioral archetypes, and create data-grounded Mavera personas

Scenario

People who reply, quote-tweet, and mention your brand reveal interests and engagement patterns. This job collects that data, analyzes behavioral patterns via Mave, and creates data-grounded persona profiles — built from real interactions, not assumptions.

Architecture

Code

Python

import os, requests, time

X = os.environ["X_BEARER_TOKEN"]; MV = os.environ["MAVERA_API_KEY"]
X_BASE = "https://api.x.com/2"; MV_BASE = "https://app.mavera.io/api/v1"
X_H = {"Authorization": f"Bearer {X}"}
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}

BRAND = "yourbrand"
queries = [f"@{BRAND} -is:retweet", f"url:x.com/{BRAND} is:quote"]

# 1. Collect interactions
interactions = []
for q in queries:
    nt = None
    for _ in range(3):
        params = {"query": q, "max_results": 100, "tweet.fields": "public_metrics,author_id",
            "expansions": "author_id", "user.fields": "username,description,public_metrics"}
        if nt: params["next_token"] = nt
        r = requests.get(f"{X_BASE}/tweets/search/recent", headers=X_H, params=params)
        if r.status_code == 429:
            time.sleep(int(r.headers.get("x-rate-limit-reset", time.time()+60)) - int(time.time()))
            r = requests.get(f"{X_BASE}/tweets/search/recent", headers=X_H, params=params)
        r.raise_for_status(); data = r.json()
        users = {u["id"]: u for u in data.get("includes",{}).get("users",[])}
        for t in data.get("data",[]):
            a = users.get(t.get("author_id"),{})
            interactions.append({"text": t["text"], "type": "quote" if "is:quote" in q else "reply",
                "username": a.get("username",""), "bio": a.get("description",""),
                "followers": a.get("public_metrics",{}).get("followers_count",0)})
        nt = data.get("meta",{}).get("next_token")
        if not nt: break
        time.sleep(1)

# 2. Pattern analysis
block = "\n\n".join(f"@{i['username']} ({i['followers']:,} fol) [{i['type']}]\nBio: {i['bio'][:120]}\n{i['text'][:200]}"
    for i in sorted(interactions, key=lambda x: -x["followers"])[:40])

analysis = requests.post(f"{MV_BASE}/mave/chat", headers=MV_H, json={
    "message": f"Analyze reply/quote patterns around @{BRAND}. {len(interactions)} interactions.\n\n{block}\n\n"
        "Identify 4-6 persona archetypes. For each: name, interaction pattern, topics, language, estimated %, value to brand."
}).json()
archetypes = analysis.get("content","")
print(archetypes[:1200])

# 3. Create personas
blocks = archetypes.split("##")[1:] if "##" in archetypes else [archetypes]
for b in blocks[:6]:
    lines = b.strip().split("\n")
    name = lines[0].strip().strip("#").strip() if lines else "X Persona"
    p = requests.post(f"{MV_BASE}/personas", headers=MV_H, json={
        "name": f"X Audience: {name}", "description": "\n".join(lines[1:])[:500],
        "psychographic": {"source": "x_reply_mining", "brand": BRAND},
    }).json()
    print(f"  Created: {name} — {p['id']}")
    time.sleep(0.3)

Example Output

## Industry Thought Leader (12%) — Quote-tweets with commentary. Influencer.
## Help-Seeker (35%) — "How do I..." replies. Active customer, upsell target.
## Lurker-Turned-Engager (20%) — Silent liker, replies to pain-point tweets. Prospect.
## Competitive Comparer (15%) — Mentions competitors alongside you. Evaluating.

Created: Industry Thought Leader — per_x_tl_01
Created: Help-Seeker — per_x_hs_02
Created: Lurker-Turned-Engager — per_x_le_03
Created: Competitive Comparer — per_x_cc_04

Error Handling

Quote tweet syntax

Use url:x.com/{username} is:quote to find quotes. If sparse, broaden to @{username} is:quote.

Archetype parsing

Code splits Mave output on ## headers. If Mave uses a different format, fewer personas are created. Review raw analysis.

---

X / Twitter Integration

Personas