Complete guide
This page is the one-stop overview. Each section links to a detailed reference page.
How it works
GPTSheetsPlus is a Google Sheets add-on. It adds custom functions you type into cells, exactly like =SUM(). Each function calls an AI or data provider and writes the answer into the cell.
- AI models: OpenAI (GPT-6 and GPT-5.6 families), Google Gemini (Gemini 4, 3.8 Flash, 3.1 Pro) and Perplexity (Agent API with live web search).
- Web scraping: Bright Data Web Unlocker turns any URL into markdown, HTML, title, description, keywords, language, links or an HTTP status code. Firecrawl is available as an alternative with your own key.
- Web search: Exa search returns URLs, domains or page text straight into your sheet.
- AI utilities: ready-made helpers for extracting a field, summarizing to bullets and generating lists.
You get a free trial on our keys. After that you buy a one-time $19 license and use your own provider keys. See Getting started.
Cheat sheet
Square brackets mark optional arguments.
OpenAI
| Formula | Model |
|---|---|
=GPT(prompt, [system_prompt], [ignore_cache]) | gpt-6.1-terra (default) |
=GPT6(prompt, [system_prompt], [ignore_cache]) | gpt-6-astra, most capable |
=GPT61(prompt, [system_prompt], [ignore_cache]) | gpt-6.1-sol |
=GPT61mini(prompt, [system_prompt], [ignore_cache]) | gpt-6.1-terra |
=GPT6mini(prompt, [system_prompt], [ignore_cache]) | gpt-6-luna, cheapest |
=GPT56(prompt, [system_prompt], [ignore_cache]) | gpt-5.6-sol |
=GPT56mini(prompt, [system_prompt], [ignore_cache]) | gpt-5.6-terra |
=OpenAI(prompt, [system_prompt], [model_name], [ignore_cache]) | any OpenAI model |
Legacy names (=GPT5, =GPT5mini, =GPT4o, =GPT41, =O1, =O1PRO, =O3, =O3MINI, =O4MINI) still work and run on current models. OpenAI reference
Google Gemini
| Formula | Model |
|---|---|
=GEMINI(prompt, [system_prompt], [model_name], [ignore_cache]) | gemini-3.8-flash (default) or any model you name |
=Gemini4(prompt, [system_prompt], [ignore_cache]) | gemini-4-argon, flagship |
=Gemini31Pro(prompt, [system_prompt], [ignore_cache]) | gemini-3.1-pro-preview |
=Gemini38Flash(prompt, [system_prompt], [ignore_cache]) | gemini-3.8-flash |
=Gemini35FlashLite(prompt, [system_prompt], [ignore_cache]) | gemini-3.5-flash-lite, fastest |
Legacy names (=Gemini25Pro, =Gemini25Flash, =Gemini20Flash, =Gemini15Pro, =Gemini15Flash) still work. Gemini reference
Perplexity
| Formula | Effort |
|---|---|
=PPLX(prompt, [system_prompt], [ignore_cache]) | fast: quick web-grounded answer |
=PPLX_PRO(prompt, [system_prompt], [ignore_cache]) | low: light multi-step research |
=PPLX_REASONING(prompt, [system_prompt], [ignore_cache]) | medium: multi-hop research |
=PPLX_AGENT(prompt, [system_prompt], [effort], [ignore_cache]) | your choice: fast, low, medium, high or xhigh |
=PERPLEXITY(), =PERPLEXITY_PRO() and =PERPLEXITY_REASONING() are aliases. Perplexity reference
Web scraping
| Formula | Returns |
|---|---|
=SCRAPE(url, [maxAge], [waitFor]) | page content as markdown |
=SCRAPE_HTML(url, [maxAge], [waitFor]) | raw HTML |
=SCRAPE_TITLE(url, [maxAge], [waitFor]) | page title |
=SCRAPE_DESCRIPTION(url, [maxAge], [waitFor]) | meta description |
=SCRAPE_KEYWORDS(url, [maxAge], [waitFor]) | meta keywords |
=SCRAPE_LANGUAGE(url, [maxAge], [waitFor]) | page language |
=SCRAPE_LINKS(url, [maxAge], [waitFor]) | all links as a JSON array |
=SCRAPE_STATUS(url, [maxAge], [waitFor]) | HTTP status code |
=BRIGHTDATA(url, [output], [render], [country]) | markdown or HTML with full control |
=Firecrawl(url, [maxAge], [waitFor]) | markdown via your own Firecrawl key |
Any waitFor greater than 0 forces browser rendering. Scrapes are cached for 6 hours. Scraping reference
Web search
| Formula | Returns |
|---|---|
=SERP(query, [number_of_result], [category]) | list of URLs |
=SERP_DOMAIN(query, [number_of_result], [category]) | list of root domains |
=SERP_TEXT_RESULT(query, [number_of_result], [category]) | list of page texts, up to 10,000 characters each |
=SERP_COMPANY(query, [number_of_result]) | list of company website URLs |
Results spill down the column. Up to 100 results. Categories: company, people, news, publication, personal site, financial report. Search reference
AI utilities
| Formula | Returns |
|---|---|
=EXTRACT_INFO(text, what_to_extract, [format]) | one field pulled out of the text |
=SUMMARIZE_BULLETS(text) | bullet-point summary (=SUMMARIZE_BULLET() is an alias) |
=LIST_STUFF(noun, [number]) | a list of items, spilled down the column |
These run on OpenAI. Utilities reference
Building a pipeline
The functions are designed to chain. A typical research sheet looks like this:
| Column | Formula | What you get |
|---|---|---|
| A | =SERP("CRM startups", 10, "company") | ten company URLs |
| B | =SCRAPE(A2) | the page content |
| C | =EXTRACT_INFO(B2, "founder name") | one clean field |
| D | =PPLX("Latest news about " & A2) | live context with citations |
| E | =SUMMARIZE_BULLETS(B2) | a short summary |
Fill A once, then drag B to E down the column. More patterns are in Use cases.
Caching, in short
AI responses are cached by prompt so that recalculation does not re-call the API. Pass TRUE as the ignore_cache argument to force a fresh answer. Scrapes are cached for 6 hours. Search and the AI utilities are not cached. Full details: Caching and recalculation.
Good habits
- Paste as values when done. Select the results, copy, then Edit → Paste special → Values only. That freezes the answers so a future recalculation cannot change or re-bill them.
- Work in batches. Google Sheets runs custom functions with a time limit. Fill a few hundred rows at a time rather than thousands at once.
- Keep prompts specific. Put the instruction in
system_promptand the data inprompt:=GPT(A2, "Reply with only the company's industry, two words max"). - Use the cheapest model that works.
=GPT6mini()or=Gemini35FlashLite()for classification and extraction,=GPT6()or=Gemini4()for writing and reasoning.
Help
- Troubleshooting explains every error message.
- Support: email us at [email protected].