DATED EXAMPLE
Why does the same keyword produce different queries?
Yes. These two US runs were about 22 seconds apart. Both kept the same four themes, but three query strings changed wording and the run-level source domains were not identical.
Use a fixed comparison rule so a wording change is not mistaken for a new topic or long-term trend. Understand repeat-run and locale variation.
Fixed protocol
- Provider
- OpenAI API
- Model
- GPT-5.6 Luna
- Tool
- web_search · low context
- Limits
- 8 searches · 500 output tokens
- Provider storage
- store: false
- Version
- native-fanout-tool/1.0.0
Observed queries
Original query strings from the API response. These are not post-run search ideas.
- best SEO tools 2026 US market SEO software comparisons official pricing features
- Semrush SEO toolkit official features pricing
- Ahrefs SEO tools official features pricing
- Google Search Console official SEO tools
Source domains on the search action — not mapped to individual queries
- ahrefs.com
- semrush.com
- help.ahrefs.com
- support.google.com
- semrush.ebundletools.com
- static.semrush.com
- images.g2crowd.com
- searchenginejournal.s3.us-west-1.amazonaws.com
- reddit.com
- youtube.com
- seo.com
- Input
- 12,905 Tokens
- Output
- 500 Tokens
- Estimate
- $0.0132
- best SEO tools 2026 US market comparison official pricing features
- Semrush SEO toolkit official features pricing
- Ahrefs official SEO tools features pricing
- Google Search Console official features SEO
Source domains on the search action — not mapped to individual queries
- ahrefs.com
- semrush.com
- help.ahrefs.com
- support.google.com
- developers.google.com
- static.semrush.com
- images.g2crowd.com
- reddit.com
- searchenginejournal.s3.us-west-1.amazonaws.com
- youtube.com
- Input
- 12,876 Tokens
- Output
- 500 Tokens
- Estimate
- $0.0132
Operator reading
This section interprets the observation. It is not a field returned by the provider.
- The high-level branches stayed stable: comparison, Semrush, Ahrefs and Search Console.
- Exact wording changed in three of four strings.
- developers.google.com appeared in the repeat, while seo.com appeared only in the first run.
What it does not prove
- This demonstrates variation, not a long-term trend.
- Exact-string comparison understates semantic similarity.
This example is not an independent model benchmark. It documents individual API observations under a published protocol and does not reproduce the ChatGPT or Gemini consumer interface.
View the planned monthly topic set
Test the workflow yourself
The observations were collected on 27 August 2026. These instructions were added on 9 October 2026. A new run can expose different queries and sources.
- Choose best SEO tools, OpenAI, English and United States. Record the exact model.
- Repeat with identical visible settings, save both observations locally and keep their timestamps.
- Compare exact query strings, then review meaning manually. Report the observed difference without estimating a general variation rate from this pair.
A related control case: How much does country change fanout queries?.