Someone says a competitor cut its price. The product team opens the site and finds that the only change was an annual-billing message. Marketing has a screenshot, but three months later nobody knows which region or plan it showed. Pricing research is not mainly a search problem; it is a record-keeping problem.
AI is useful for extracting plans, allowances and limits from long pages, highlighting differences between snapshots, and organising raw text into a table. It should not be the system that declares a competitor “cheaper.” A missed billing unit, tax rule, promotion or currency can reverse the conclusion.
Define the comparison before choosing the tool
A durable price ledger has four layers. Capture page facts: URL, date, region, currency and tax treatment. Capture plan structure: plan name, monthly or annual billing, displayed price and billing unit. Capture constraints: seats, usage, feature gates, overages and trial terms. Only then write a business interpretation.
Do not record “Pro: $99/month” alone. Record whether it is per user, per workspace, annual-equivalent, the included seats, the overage rule and the market shown. The same number can mean an individual subscription, a minimum team commitment, or a first-month promotion.
Save a page snapshot, not only a screenshot
Save the public URL, a PDF or complete screenshot, and copyable page text each time. Do not ask an AI tool or crawler to bypass a login for a quote; mark “contact sales” as a known state. Monthly checks are usually enough for mature SaaS products, weekly checks may suit promotion-heavy categories, and every record needs a date.
Ask AI to extract facts first, then compare
From this public pricing-page text, create a table with plan name, displayed price, billing cycle, currency, billing unit, included seats or allowance, overage charges, trial/promotion, and explicitly stated feature limits. Mark missing fields “not disclosed.” Add a short source excerpt for every row. Do not convert currency or judge value.
Then compare the old and new tables:
List only source-supported changes in plans, displayed prices, cycles, allowances, seats, feature gates, trials and promotions. Put wording changes with uncertain meaning under “human review needed.” Do not call anything a price increase or cut unless the billing unit, currency, cycle and included content are the same.
This makes the model’s work auditable. It removes repeated reading without pretending to be a pricing strategist.
Four common false alarms
- Monthly and annual billing: a lower annual equivalent does not mean the monthly price fell.
- Renamed plans: compare allowances, audience and features before treating a new name as a new product.
- Reduced free access: a fixed paid price can still mean a higher entry barrier when trials or exports shrink.
- Regions and taxes: the same site can show different currency, VAT and promotion messages. Fix the browsing region and record it.
Turn a confirmed change into useful work
A price ledger should not become a wider spreadsheet every month. After confirmation, write a short impact note: does it affect new or existing customers, lower a trial barrier, or raise enterprise spend? AI can draft this note, but a person who knows the market must judge intent.
- Fact: On July 10, the competitor increased included team seats from three to five; annual price stayed the same.
- Hypothesis: It may improve its initial comparison for teams of five or fewer.
- Verify: Does it apply in every region and change overage pricing?
- Action: Update sales comparison language and test the issue in customer interviews.
Give spreadsheets, search and automation different jobs
Google Sheets or Notion can hold the fact table. ChatGPT, Claude or Gemini can extract and draft differences. Perplexity can help locate official pricing pages, announcements and help-centre sources, but a search summary is not price evidence. Confirm the current official page or announcement.
Once the set grows beyond a dozen competitors, an n8n or Zapier reminder can flag page changes and send old/new text to an AI for a review queue. Stop before writing to the master ledger: temporary A/B tests and cookie-based copy otherwise contaminate history.
Pre-publication checklist
- Does every change have a date, URL and source excerpt?
- Are monthly, annual, promotional, tax and regional conditions separated?
- Has “not disclosed” been incorrectly converted to zero or unlimited?
- Did a person review material plan changes?
- Are you using only public information, not restricted quotes or pages?
For source finding, pair this with the Perplexity competitor research checklist. The goal is not for AI to tell you who is cheaper; it is to help your team preserve evidence, explain impact, and act when a real change happens.