AgentSEO research & benchmarks

Inspect the evidence.
Then try your workflow.

How much smaller is the payload? What does a search request return? Explore our measurements, inspect recorded API behavior, and build a request for your own use case.

First-party research by AgentSEO. Methods, samples, and limits are part of each finding.

01 / Payload benchmark / July 12, 2026

See what sits behind the average.

The enriched output was 72.6% smaller on average across three queries. The individual reductions vary. Switch queries below to see the difference.

This compares provider task-result arrays with AgentSEO’s normalized and enriched JSON. It measures bytes, with the provider’s outer response wrapper excluded. The July study is historical; it has not been silently replaced with today’s results.

Compare provider results with

Enriched output includes normalized JSON, agent_support, and markdown_summary.

Payload size · UTF-8 bytes of minified JSON

seo api

Provider32,265
Enriched6,166

rank tracking api

Provider39,897
Enriched6,525

content gap analysis api

Provider13,087
Enriched6,102

72.6%

Average per-query byte reduction across these three queries.

This measures payload bytes, not tokens, API cost, or response speed.

See all measurements and calculation
July 12, 2026 · all three original observations
QueryProvider bytesCore bytesEnriched bytesEnriched reduction
seo api32,2652,5766,16680.9%
rank tracking api39,8972,6976,52583.6%
content gap analysis api13,0872,5386,10253.4%

Per-query reduction = (1 − output bytes ÷ provider bytes) × 100. The headline is the mean of those three percentages, not the reduction in the combined byte total. Provider response wrappers are excluded.

Three selected queries are a small directional sample. Smaller responses do not establish token savings, lower bills, faster responses, or better rankings for your workflow.

02 / Production observations / September 20, 2026

From a request to usable fields.

We submitted three search requests to AgentSEO’s production API and polled their jobs. All three completed in that session. Select a query and step through the recorded stages.

A concrete integration observation

Each request asked for up to 10 results. We observed 8, 9, and 9 returned items. Build for the returned array length, rather than assuming the requested limit will be filled.

Saved production observations · September 20, 2026. These controls make no API calls.

Observed status: completed

What came back for “seo api

Result bytes
6,156
Returned items
8
SERP feature records
3

items[]

Observed fields: url, rank, title, description.

agent_support

Summary, evidence, confidence, limitations, and recommended next calls.

markdown_summary

2,417 UTF-8 bytes within the result, for readable context.

search_parameters

United States; language en; explicit location code 2840.

This is an inventory of the recorded response. Full result content is not reproduced.

Compare all three observations
September 20, 2026 · one session · limit 10 · United States (2840)
QueryJob statusResult bytesItemsFeature records
seo apicompleted6,15683
rank tracking apicompleted7,28992
content gap analysis apicompleted7,29293

How we measured

Each query was submitted once to POST /api/v1/search with limit 10 and location_code 2840. We measured UTF-8 bytes of the minified completed job’s result object, excluding the job envelope, headers, and compression.

What this cannot tell you

This is not an uptime, latency, or accuracy benchmark. Caching and upstream behavior were not isolated. Today’s measurements have no raw-provider comparison and cannot refresh the July reduction claim.

Published data contains aggregate sizes and field inventories. Keys, job IDs, account details, and full search-result content are excluded. Running the script requires your API key and uses credits.

03 / Your evaluation

Try a question from your own work.

A sample can explain the response. Your own query tells you whether it fits the workflow you are building.

This preview creates a search request for the United States. It does not generate pretend search results. Run it in your environment or use the Playground to inspect a real response.

Check plans and credit costs

Build a request preview. Nothing is sent and no credits are used here. Running it with your key uses your plan’s credits.

curl https://www.agentseo.dev/api/v1/search \
  -H "x-api-key: $AGENTSEO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"seo automation tools","limit":10,"location_code":2840}'

Set AGENTSEO_API_KEY in your environment. Poll the returned jobId until completed or failed; a queued response is not the search result.

Evaluate the evidence

Keep the method with the claim.

These are measurements of our own product. We publish their scope so you can challenge the interpretation and repeat the checks.

Reproduce the inputs

Start with the published request, sample size, date, and measurement boundary.

Check the right outcome

A compact result and a completed job do not prove that the recommendations will improve rankings.