# Deep research in ChatGPT

How ChatGPT deep research and company knowledge use Load Factor's search and fetch tools, and the record ids fetch accepts.

ChatGPT's deep research and company knowledge work through two Load Factor tools: `search` finds records and `fetch` opens them in full. You do not call them yourself, but knowing how they behave helps you phrase research requests and read the citations. Any other client can use the same two tools.

## How it works

1. **`search`** takes free text and returns up to 22 typed results, each with an `id`, a `title` and a `url`: queue projects, companies, EIA power plants, deals and news, incentive programs, and up to two Load Factor Insights articles.
2. **`fetch`** takes one of those ids and returns the full record: a readable dossier (`text`), a `url` to cite, and `metadata`.

A report built this way cites Load Factor links, so you can open each record and check it.

## Record ids

`fetch` accepts the ids `search` returns, or ones you build by hand:

| Id | Record | Example |
|---|---|---|
| `project:<queue id>` | An interconnection-queue project | `project:J1229` |
| `company:<exact name>` | A company, by one exact name variant | `company:Apex Clean Energy` |
| `company:cik:<SEC CIK>` | An SEC registrant reviewed by Load Factor, with the Load Factor profile of the names linked to it ([SEC company intelligence](https://loadfactorenergy.com/developers/mcp/sec-company-intelligence)) | `company:cik:753308` |
| `plant:<EIA plant id>` | An EIA power plant | `plant:57695` |
| `news:<story id>` | A news story | `news:381b0598bb1988f6` |
| `deal:<deal id>` | A curated deal record | `deal:ma-2025-026` |
| `insight:<id>` | A third-party deal-analysis headline (title and link) | an id returned by `search` |
| `program:<DSIRE program id>` | An incentive or policy program | `program:22803` |
| `dg:<NY DG project id>` | A New York distributed-generation application | `dg:NG-624906` |
| `article:<slug>` | A Load Factor Insights article, in full | `article:ercot-queue-who-is-building-texas` |

The other tools take the bare value instead: `get_project` takes `J1229`, `get_plant` takes `57695`.

## How search matches

- Exact queue ids come first, then renewable queue projects (live before withdrawn), companies, EIA plants, news and deals that name the query, and incentive programs.
- Short words (under five letters, no digits, such as AES or PPA) match whole words only.
- When the whole text matches little, the words are used separately: ISO codes, state names or codes, and fuels act as filters; capitalised words (names) must match. "NextEra solar Texas" finds NextEra's solar projects in Texas.
- Results are limited to what your Load Factor plan can see.

On plans that open full project detail only for priority projects, `fetch` returns the list-level fields for other projects.

## Tips for research requests

- Name Load Factor and the scope: "Using Load Factor, research the battery storage pipeline in ERCOT and who is building it."
- Ask for citations to Load Factor records, and for the confidence level of every developer name.
- `search` finds records; it does not count them. For totals and rankings ("how many", "top 10 by MW"), use a regular chat with Load Factor chosen, where the assistant can use `queue_stats` and the other tools.
