Benchmark 01 / July 12, 2026
Search payload size benchmark
In this three-query run, AgentSEO normalized core JSON was 88.6% smaller on average than the provider task-result payload. The enriched agent output was 72.6% smaller on average while keeping the fields an agent needs to summarize, route, and act.
Less parsing work
The core contract removes provider noise before the response reaches an agent, script, or workflow step.
More usable outputs
The enriched response adds agent_support and markdown_summary so the next action is easier to inspect or pass downstream.
A bounded claim
This proves payload-size reduction for one small SERP sample. It is not a universal latency, token, or cost claim.
Results
The provider column measures only the returned task-result array, excluding its outer response wrapper. This makes the comparison conservative. The enriched column includes both structured agent support and a Markdown summary, even though those fields repeat selected information for downstream usability.
| Query | Provider result | Normalized core | Enriched output | Core reduction | Enriched reduction |
|---|---|---|---|---|---|
| seo api | 31.5 KB | 2.5 KB | 6.0 KB | 92% | 80.9% |
| rank tracking api | 39.0 KB | 2.6 KB | 6.4 KB | 93.2% | 83.6% |
| content gap analysis api | 12.8 KB | 2.5 KB | 6.0 KB | 80.6% | 53.4% |
Method
- Query set
- seo api, rank tracking api, content gap analysis api
- Search context
- United States, English, desktop
- Result depth
- 10
- Measurement
- UTF-8 bytes of minified JSON
- Provider scope
- Provider task result array only. The provider response wrapper was excluded.
What was compared
Provider result: the task-result array returned by the provider endpoint, without the surrounding API envelope.
Normalized core: the compact SERP object used by AgentSEO, including results, detected features, search parameters, and available keyword metrics.
Enriched output: normalized core data plus the agent-support object and human-readable Markdown summary.
Limitations
- This is a small directional sample, not a guarantee for every query or endpoint.
- Payload size varies with result count, SERP features, and provider response shape.
- Latency is reported for reproducibility but is not a latency benchmark; the run was not repeated enough to support a performance claim.
- Byte reduction is not the same as token reduction. No tokenizer-specific claim is made.
- The enriched output intentionally repeats selected information in Markdown for direct use by agents and humans.
Interpretation
This run supports a narrow claim: for these three queries, the AgentSEO search contract substantially reduced JSON payload size while adding an action-oriented summary. It does not establish a universal token reduction, latency advantage, or cost advantage.
Future runs should expand the query set, repeat requests, and add tokenizer-specific measurements before making model-level token claims.
Practical next step
Use the benchmark as a method, not a promise.
The useful question for production teams is whether compact, structured outputs make a real workflow safer and cheaper to run. Start with one bundle, check credit use, and compare the agent output against the raw provider work you would otherwise maintain yourself.