Research Engine · proprietary technology
Unlimited queries. Five AI platforms. Structured answers.
The Research Engine puts your questions to ChatGPT, Claude, Gemini, Perplexity, and Grok at once, as many queries as you want to run. Each platform is forced to search the live web, take a position, and show its reasoning, and every answer comes back as normalized, structured data: the answer, the reasoning, the citations, and the source domains behind it.
Asking one platform gives you an anecdote. Asking five shows you where they agree, where they split, and which names keep surfacing no matter who you ask. That spread is the difference between an opinion and a read.
What does one query return?
One normalized record per question, per platform.
Every platform answers differently, formats differently, and cites differently. The Research Engine flattens all five into one contract, so a question asked across the field comes back as rows you can compare, store, and build on.
The answer, verbatim
What the platform actually said when asked to recommend: the full answer text, plus the winner and ranking extracted from it as structured fields.
The reasoning, forced
Every query runs under a strict structured-output contract that requires the platform to explain its call. You get the reasons behind every recommendation as data, not prose you have to mine.
The citations, grounded
Every source the platform's own live web search actually retrieved, normalized, deduplicated, and reduced to citation domains. A grounding gate separates real retrieved sources from URLs a model merely claims.
The run, accounted for
Each record carries its status, latency, token counts, and cost. Refusals are detected and labeled. A failed call persists as an error record instead of vanishing, so a run is always complete and auditable.
Why not just call the platforms yourself?
Because the hard part is not asking. It is trusting what comes back.
We evaluated the off-the-shelf response aggregators and built direct integrations instead: the aggregators cap prompt length, return unstructured text, cannot guarantee reasoning, and cannot force web search on every platform. The method that makes the data trustworthy is exactly what they cannot carry.
01
Web search, enforced per platform
Each of the five platforms has its own integration with live web search forced on, with grounding prompts tuned per vendor. Answers reflect what the platform finds today, not what it memorized in training.
02
A strict structured-data contract
Answers come back as structured JSON with the reasoning required, not free text. When a platform emits broken JSON, and they do, a deterministic repair pass recovers it without touching the content.
03
A vendor-aware grounding gate
Citation handling is built per platform, because each one reports its sources differently. URLs are normalized and deduplicated, reduced to registrable domains, and only counted as grounding when the platform's own search retrieved them.
04
Reliability you can build on
A query never crashes a run. Refusals, truncations, and vendor errors each degrade to a labeled record, so you always get one row per question per platform, with honest status on every one.
What do we do with the data?
Measure the field, map the language, then close the gaps.
Inside our own engagements, the Research Engine is the read behind everything else. It measures who each platform recommends in your category and why, normalizes the reasons into themes, correlates what the winners share, and maps search intent to the language AI answers actually use. Those findings aim our editorial services at the exact gaps the read exposed.
The measurement stays honest on purpose. We read what the platforms say when asked to recommend, not language we coached into them, so the picture reflects the market’s view rather than your own marketing.
What does one run look like, end to end?
Queries in. Five platforms verified. Records out.
Queries in
General intent
no brand named · the discovery read
Head-to-head
forced pick · the comparison read
Open research
any question, flow-through
+ required and optional variables shape each run
5 platforms · queried at once
ChatGPT
Claude
Perplexity
Gemini
Grok
live web search forced on, every platform
Every answer, verified
Reasoning, required
strict structured-JSON contract
Grounding gate
only retrieved sources count
Deterministic parse
JSON repair, byte-stable
Refusal detection
a dodge is labeled, not counted
One JSON record · per query × platform
The answer
verbatim + winner · ranking
The reasoning
why, as data
Citations
normalized · deduplicated
Grounded domains
vendor-retrieved only
Run accounting
status · latency · tokens · cost
Always-write: errors, refusals, and truncations come back as labeled records, never gaps
How is a query shaped?
Three ways to ask. One contract back.
Every call is a query plus the required and optional variables that shape the run. Whatever the mode, what comes back is the same normalized JSON record per platform.
01
General intent
Buyer-style questions that never name a brand, so the answer shows the unprompted field: who surfaces, who wins, and the reasons each platform gives. This is the discovery read.
02
Head-to-head
Name the contenders and force a choice. Each platform must pick, rank, and explain, so you measure preference directly instead of inferring it. This is the comparison read.
03
Open research
The flow-through mode: any question you want answered across all five platforms at once, with the same grounded, structured record back. A research tool in its own right.
The engine is tuned to extract opinions with evidence attached. Output quality follows query quality: query design is the craft.
Can I access it directly?
The engine runs today. The API is opening up.
The Research Engine is in production now in Austin, Texas, running the research behind every Trinzik engagement. Direct API access is opening to early partners: you send the questions, we orchestrate the five-platform queries inside the same structured setup we use ourselves, and you get the normalized records back. What you build with the data is up to you. On API access, query design is yours; inside our engagements, we do it for you.
Two things are in preparation for API partners: a login portal with everything an integration needs in one place, and a public API library documenting every call the system exposes, the required and optional variables each one takes, and the JSON output field by field. An MCP server is on the roadmap alongside them, so AI agents and tools can query the engine natively instead of through custom integration code.
See how AI recommends you today.
A consult runs your real category questions through the five-platform read and shows you the recommendations, the reasoning, and the sources behind every call, live against your own market.