Ad frequency inside ChatGPT jumped from 0.4% to 26% in roughly three weeks. If you're in B2B SaaS, your category already leads total impressions. Here's how to find out who's bidding against you. In mid-April 2026, ads appeared in about 0.4% of ChatGPT responses. By late May, that number surged to 26%. Three weeks. That's how quickly the auction scaled. For B2B SaaS companies, the kicker: B2B Software and SaaS accounts for 26.3% of total ChatGPT ad impressions, more than any other category. Your buyers are already asking ChatGPT comparison and recommendation questions that used to be on Google. A competitor's sponsored card sits below the answer, clearly labeled "Sponsored," with a headline, short description, and destination URL. There is no ad library for ChatGPT, nor any equivalent of Meta's or Google's transparency tools. To know who's buying attention in your category, you must investigate.

Why This Deserves Ops-Grade Attention Now

The scale alone justifies the work. Ads now appear in about 49% of US ChatGPT replies and 76.4% of shopping-related queries. By June 20, 2026, 1,795 advertisers were running placements, a 300x increase from the prior week. OpenAI dropped the $50,000 minimum spend on May 5 and launched a self-service Ads Manager, collapsing the barrier to entry. Here's what should concern demand gen operators most: when an ad appears, the advertiser is mentioned in the organic answer only 5.0% of the time. Competitors can buy adjacency to recommendations without earning them. Approximately 20% of ChatGPT queries carry direct commercial intent, indicating a significant high-intent window. Ignoring this means you'll learn about competitor activity from a pipeline gap, not from a dashboard—a lagging indicator when you need a leading one.

Build a Prompt List That Mirrors Real Buyer Language

Start with 30 to 50 prompts. Don't guess. Pull from converting paid search terms, branded organic queries, on-site search logs, sales call transcripts, and support tickets. Rewrite them in the conversational format people use with ChatGPT: full sentences, context, constraints. Cover these categories: If competitors bid on prompts you haven't mapped, you won't see them. Your prompt list is your ad library.

Run the Prompts, Capture Four Data Points

Don't run each prompt once. ChatGPT's ad auction rotates advertisers based on bid, relevance signals, and session variables. Plan for 20 to 30 runs per prompt across multiple days, varying the time of day and clearing cookies between batches. For every sponsored card that appears, capture four fields:
  1. Ad title: exact headline copy, character for character. Headlines change quickly.
  2. Ad description: the body text (averaging 19 words, but varies).
  3. Final URL: strip UTMs for the canonical landing page, but keep the full URL in a secondary column for tracking-pattern analysis.
  4. Impression share: calculated, not observed. Count how many times each advertiser appeared out of total runs per prompt. If Competitor A shows in 12 of 25 runs, that's 48% impression share for that prompt.
Tag each row with the prompt, date, time, and session type (Free vs. Go tier, geo). Set up the spreadsheet to pivot by prompt, competitor, and week. Ad copy iterates quickly: the same advertiser might rotate three or four headlines in a week. Capturing only the headline misses the iteration pattern, which reveals more about strategy than any single creative.

Cadence That Produces Signal, Not Noise

A one-day snapshot will mislead you. You'll catch whoever won the auction that morning and miss the rotation happening every other day. Minimum cadence for real signal: This is manual and tedious. The upside: with only 1,795 advertisers in the system and ad relevance still inconsistent (50.1% of ads are on-topic, 21.6% have no meaningful connection to the query), the competitive intelligence you gather now has outsized value. Categories aren't saturated, and positioning data is scarce. Teams doing this work today are building a baseline everyone else will wish they had in Q4. One more thing worth weighing: experts recommend pairing paid monitoring with Generative Engine Optimization, structuring your content (FAQs, comparison pages, machine-readable specs) so AI engines understand and cite your brand organically. With only 5% alignment between advertiser and organic answer, brands appearing in both the recommendation and the sponsored card will own the highest share of trust in these conversations. Back in April, 0.4% of responses carried an ad. Now it's 26% and climbing. The auction is live. The question isn't whether competitors are bidding on your category's prompts; it's whether you'll know about it before your pipeline tells you.