NegativeIQ

AI-powered search term classification for Google Ads.

Free. Open source. No agency required.

---

What is NegativeIQ?

Most Google Ads accounts are silently leaking budget every day — on job seekers, competitors, and irrelevant queries that will never convert. Manually reviewing search terms is slow, inconsistent, and doesn't scale.

NegativeIQ automates the entire process.

It fetches your search terms from Google Ads into a Google Sheet, classifies each one using AI, and lets you push confirmed negatives back into your campaigns with a single click. No data leaves your account. No subscriptions. No black box.

Fetch → Pre-classify → AI-classify → You review → Push negatives → System learns

---

How it works

Step	What happens
1. Fetcher runs daily	Pulls all search terms from Search + PMax campaigns into your Google Sheet
2. Pre-classify	Rule-based pass against your keyword lists — zero API cost
3. AI-classify	Remaining terms sent to AI in batches — classified in seconds
4. You review	Check classification, confidence score, and AI reasoning. Override if needed
5. Export	Validated negatives written to staging sheet
6. Uploader runs	Books negatives into your campaigns or shared negative lists
7. Learning loop	Your corrections feed back into keyword lists and AI examples

---

Architecture

Google Ads
    │
    ├── Fetcher Script (Google Ads Scripts — runs daily)
    │       → Writes to Google Sheet: SearchTerms + PMax SearchTerms
    │
Google Sheet (Apps Script)
    ├── AI Config        — client profile, categories, AI examples, neg_lists registry
    ├── Keyword Lists    — Brand Pure / Brand Kombi / Bestand / Sperrbegriffe / Mitbewerber / Brand Noise
    ├── SearchTerms      — standard search campaign terms
    ├── PMax SearchTerms — Performance Max category label insights
    ├── Negative_Export  — staging area for the uploader
    └── Expansion_Ideen  — keywords flagged as expansion opportunities
    │
    └── Uploader Script (Google Ads Scripts)
            Books Negative_Export rows into Google Ads
            Deletes successfully uploaded rows from the sheet

---

Files

File	Where it lives	What it does
AI_Max_NegativeIQ_V2_3d___GAS.js	Google Sheet → Extensions → Apps Script	Main Apps Script — classification, export, learning loop
AI_Max_Negative_IQ_v2_3_fetcher.js	Google Ads → Tools → Scripts	Fetches search terms + PMax insights into the Sheet
AI_Max_Negative_IQ_v3.js	Google Ads → Tools → Scripts	Reads Negative_Export and books negatives into Google Ads
AI_Max_NegativeIQ_V2.3b_Template.xlsx	—	Sheet template reference

Active version: V2_3d

---

Setup

1. Google Sheet

Create a new Google Sheet with these tabs (exact names, case-sensitive):

SearchTerms · PMax SearchTerms · AI Config · Keyword Lists · Negative_Export · Expansion_Ideen

Or copy the template directly →

2. Apps Script

Open the Sheet → Extensions → Apps Script. Paste AI_Max_NegativeIQ_V2_3d___GAS.js and save.

Set your API key: gear icon → Script Properties → add:

Key:   AIML_API_KEY
Value: your-api-key-here

Uses AIML API with gpt-4o-mini by default. Swap model or endpoint in callAIBatch() if needed.

3. AI Config sheet

A1:B7 — Client Profile

Key	Value
client_name	Your client or account name
client_industry	e.g. E-Commerce, SaaS, Versicherung
default_neg_list	Name of your default shared negative list in Google Ads
brand_noise_list	Name of your Brand Noise negative list
conv_protect	TRUE to skip terms with conversions
confidence_warn	Confidence threshold for warnings, e.g. 0.7
batch_size	Terms per API call, e.g. 15

A9:B30 — neg_lists registry (one row per shared negative list, key neg_lists)

E1:G50 — Categories

Code	Description	Action
Brand Pure	Exact brand name or clear purchase intent	Review
Brand Kombi	Brand + geo/product, direct intent	Review
Brand Noise	Brand + research/dissatisfaction signal	Negativ
Mitbewerber	Competitor brand terms	Negativ
Junk	Irrelevant, no intent	Negativ
Informational	Research queries, no purchase intent	Negativ
Informational-Potential	Research with SEO opportunity	Expansion
Bestand (Aktiv)	Already covered by active keywords	Review
Unklar	AI could not classify confidently	Review

H1:I50 — AI Examples (seed examples to guide the AI)

4. Keyword Lists sheet

Headers in row 3, data from row 4:

A	B	C	D	E	F
Brand Pure	Brand Kombi	Bestand (Aktiv)	Sperrbegriffe	Mitbewerber	Brand Noise Terms

Brand Noise Terms (col F) are noise modifiers: erfahrungen, alternative, kündigen, test, probleme, fake — not full terms.

5. Fetcher Script

Google Ads → Tools → Bulk Actions → Scripts → New script → paste AI_Max_Negative_IQ_v2_3_fetcher.js

Update the config block:

const SPREADSHEET_URL = 'YOUR_SHEET_URL_HERE';
const LOOKBACK_DAYS   = 'LAST_30_DAYS';
const MIN_CLICKS      = 1;
const PMAX_MIN_IMPR   = 0;

Set to run daily.

6. Uploader Script

Google Ads → Tools → Bulk Actions → Scripts → New script → paste AI_Max_Negative_IQ_v3.js

const SPREADSHEET_URL   = 'YOUR_SHEET_URL_HERE';
const EXPORT_SHEET_NAME = 'Negative_Export';

Run manually after each export cycle, or schedule it.

---

Pre-Classification Logic (Brand Tier)

Term contains brand name?
│
├── YES — Exact brand name only?           → Brand Pure (confidence 1.0)
├── YES — Brand + known noise modifier?   → Brand Noise → routes to brand_noise_list
├── YES — Exact match in Brand Kombi list? → Brand Kombi (confidence 1.0)
├── YES — Ambiguous modifier?             → AI Queue (AI decides: Kombi vs Noise)
└── NO  — Other list checks:
          Sperrbegriffe → Junk
          Bestand exact match → Bestand (Aktiv)
          Mitbewerber → Mitbewerber
          No match → AI Queue

Conv-Schutz runs first. If a term has conversions and conv_protect = TRUE, it is flagged ⚠️ Conv-Schutz and skipped during export regardless of classification.

---

Workflow

[Fetcher — daily]       Appends new search terms to both term sheets

[Sheet Menu]
  1. Prepare Layout     Styles sheets, sets up dropdowns
  2a. Pre-Classify      Config-list matching (no API cost)
  2b. AI-Classify       LLM classification for AI Queue rows only
  → Review col G (Classification), col H (Confidence), col I (Reason)
  → Override in col J (Korrektur)
  → Check col K (Validiert ✅) for rows ready to export
  4. Export             Writes to Negative_Export + Expansion_Ideen
  5. Update Config      Learning loop → updates Keyword Lists + AI examples

[Uploader]              Books Negative_Export into Google Ads, deletes uploaded rows

---

Diagnose & Debug

Run 🔍 Diagnose & Debug from the sheet menu for a full status report:

• Sheet tab presence · Column header validation · Classification distribution
• AI Config summary · neg_lists registry · Category list · Keyword Lists counts · API key check

---

Changelog

V2.3 (current)

• Brand tier split — Eigene Marke replaced with three granular categories: Brand Pure, Brand Kombi, Brand Noise
• Keyword Lists expanded to 6 columns
• Pre-classify brand logic updated with noise modifier routing
• AI system prompt updated with Brand tier rules
• Match type defaults: Mitbewerber + Brand Noise both default to Exact
• Export: Brand Noise auto-routes to brand_noise_list
• Learning loop: Brand tier corrections write to correct columns

V2.2

• Config sheet split into AI Config and Keyword Lists
• getConfig(), learningLoop(), debugDiagnose() updated accordingly

V2.1 / V2.1.1

• PMax SearchTerms sheet added
• Fetcher extended with fetchPMaxSearchTerms()
• neg_lists registry added to AI Config
• Mitbewerber match type: Phrase → Exact

V2.0

• Two-step classification: Pre-Classify → AI-Classify
• Conv-Schutz introduced
• Learning Loop introduced

V1

• Initial uploader: reads Negative_Export, books negatives, deletes uploaded rows

---

Notes

• List names are case-sensitive in Google Ads. Check spelling before running the uploader.
• Brand_Noise_Negatives list must be created manually in Google Ads first.
• The fetcher aggregates metrics for duplicate terms instead of creating new rows.
• Learning Loop only processes rows with Status = Verarbeitet and a value in col J.
• AI uses gpt-4o-mini via AIML API at temperature: 0.1 for consistent output.

---

License

MIT — free to use, modify, and share.

---

If NegativeIQ saves your account money, feel free to feed the capybara with a few API tokens 🐾

_{Built by a Google Ads specialist who got tired of manual negative reviews.}