And the add-on can do all that inside<\/strong><\/em>\u00a0Google Sheets<\/strong>, making it extra convenient to use your results for cold outreach campaigns with GMass<\/strong>.<\/p>\n Note: If you want a guide to installing GPT for Sheets or figuring out its pricing structure, check out our <\/strong><\/em>complete guide to GPT for Sheets<\/a><\/strong><\/em>\u00a0\u2014 then come on back to this article.<\/strong><\/em><\/p>\n Here are the results of (and instructions for) different prospect hunting experiments I ran with GPT for Sheets.<\/p>\n You should be able to replicate these all pretty easily in your own Google Sheets.<\/p>\n If you have a list of websites\u2019 contact pages and you want to find an email address on each one for your campaign, GPT for Sheets can handle that legwork.<\/p>\n In this example, I am using two GPT for Sheets formulas to find email addresses for 10 American-based sock manufacturing companies.<\/p>\n The first is GPT_SCRAPE, which grabs the entire contents of a web page<\/strong>\u00a0and copies it as text into a Google Sheets cell.<\/p>\n The second is GPT_EXTRACT, which sifts through a Google Sheets cell looking for a specific type of data<\/strong>\u00a0(in this case, an email address).<\/p>\n Important: In order to use the GPT_SCRAPE function, you need to first follow GPT for Work\u2019s <\/strong><\/em>instructions<\/a><\/strong><\/em>\u00a0on installing a Google Apps Script. It only takes a couple of minutes.<\/strong><\/em><\/p>\n Once you follow the instructions for getting GPT_SCRAPE working, it adds a new Scrape<\/strong>\u00a0menu to Google Sheets.<\/p>\n I used that to scrape the contents of the URLs in Column A into Column B.<\/p>\n Then I used GPT_EXTRACT to find an email address in each website\u2019s scraped results.<\/p>\n The function did a good job with this task, finding the accurate email address on the page in every case<\/strong>. (One page did not have an email address, so GPT for Sheets just left the field blank. I was glad to see it didn\u2019t hallucinate, as AI still just loves<\/em>\u00a0to do.)<\/p>\n But, as always with AI, the results are a little bit quirky. In a few cases, it doubled up the email addresses.<\/p>\n And there\u2019s no built-in setting to tell GPT for Sheets just to grab a single result. So it would take another formula to make the email column ready for a mail merge campaign.<\/p>\n Good news: This cost me less than 1 cent of my GPT for Sheets credit. GPT_SCRAPE does not use your paid GPT for Sheets tokens<\/strong>, and the email extraction didn\u2019t even knock a full penny off my total.<\/p>\n The above example worked great. However, I had to manually research all those companies\u2019 contact page URLs.<\/p>\n So I decided to try something that would require less manual work on my end<\/strong>.<\/p>\n In this case, I am only providing the <\/strong>name<\/strong><\/em>\u00a0of a company<\/strong>. And we\u2019ll use GPT for Sheets\u2019s web research function, GPT_WEB, to find what I need from there.<\/p>\n I started by putting each company\u2019s name into column A.<\/p>\n Then in column B, I concatenated together a prompt.<\/p>\n I structured my prompt not just to give me an email but also the URL. That way I could spot check the AI\u2019s results \u2014 and also, in circumstances where it didn\u2019t find a contact email, I could manually check the website myself. (That\u2019s still manual work, but if GPT for Sheets takes care of most<\/em>\u00a0of my contacts that\u2019s a lot less of it.)<\/p>\n In column C, I used this simple formula to run the prompt.<\/p>\n Here are the results.<\/p>\n I can\u2019t complain too much. It found an accurate email for 8\/10<\/strong>. It couldn\u2019t find the available contact page or email for one of them (row 10) and hallucinated on the one site that didn\u2019t have an email address on its contact page (row 6).<\/p>\n But 80% accuracy is about the trade-off I\u2019d be willing to take for this level of prospecting automation<\/strong>. (It\u2019s also, realistically, about the accuracy rate I\u2019ve seen when I\u2019ve hired virtual assistants and others to do prospecting for me manually.)<\/p>\n Now from here I\u2019ll use another GPT for Sheets formula to bring the URLs into one column and the email addresses into another column.<\/p>\n After a few unsuccessful attempts to get GPT for Sheets to split out the URL and email address using just one function, I wound up using two separate GPT_EXTRACT formulas.<\/p>\n In the Website<\/strong>\u00a0column: And in the Email<\/strong>\u00a0column: The results in both columns are perfect. However\u2026 they\u2019re perfect based on the data I provided. And, as we saw earlier, the data was 80% correct. So this is a \u201cgarbage in, garbage out\u201d situation: The better the prospecting data you have, the better the extractions will be<\/strong>.<\/p>\n Overall, I\u2019m really happy with the results I got from this approach. For not having to do any<\/em>\u00a0manual research other than knowing some companies\u2019 names, I\u2019ll take it.<\/p>\n Of course, you may be after a higher level of accuracy. I only had 10 contacts; with a larger contact list, this method could require a lot of manual spot checking and\/or filling in the gaps.<\/p>\n And one other downside: GPT_WEB is expensive. It\u2019s not billed with GPT for Sheets\u2019 usual token system. So you could find yourself burning through your GPT for Sheets budget <\/strong>really<\/strong><\/em>\u00a0fast<\/strong>\u00a0if you used this technique on a large data set.<\/p>\n Then again, based on the time it\u2019s saving and the pretty strong accuracy, it just might be worth it.<\/p>\n For my next trick\u2026<\/p>\n Let\u2019s throw something more complicated than a website at GPT for Sheets.<\/p>\n What if you have a list of prospects in a file<\/strong>, like a PDF or scanned document?<\/p>\n For those cases: There\u2019s a function in GPT for Sheets called GPT_VISION that can analyze an image. In the examples on their website, they use it to write product descriptions, but I wanted to see if it could extract data.<\/p>\n GPT_VISION needs a prompt and URL of an image<\/strong>, in the structure of Note: If you have a PDF, you\u2019ll need to convert it to an image. GPT_VISION only works with png, gif, jpg, and webp file types.<\/em><\/p>\n Here is the first image I threw at it:<\/p>\n And I used this prompt:<\/p>\n Here\u2019s the result:<\/p>\n It successfully extracted all 10 emails from the image. And now I can use another formula to get them each into their own rows. What about data that\u2019s less structured? I generated some AI pink slime and sprinkled the 10 email addresses throughout. Here\u2019s how that image looks:<\/p>\n I used the same prompt as before, and once again GPT_VISION successfully found all 10 email addresses.<\/p>\n Finally I tried the toughest test of all \u2014 a handwritten list.<\/p>\n And GPT_VISION got that perfect, though I wonder if it would struggle with sloppier handwriting. (Yes, that\u2019s a backdoor brag about my not-terrible penmanship in that image.)<\/p>\n I thought GPT_VISION would possibly burn through my GPT for Sheets budget, since it uses gpt-4o, but all of my tests above only cost around one cent<\/strong>.<\/p>\n Now that we\u2019ve done some email and URL hunting on company websites, let\u2019s scrape info about individual prospects<\/strong>.<\/p>\n For this example, I had GPT for Sheets try to research the following things about Ajay, the CEO of GMass (using these various prompts).<\/p>\n Note: Scraping of LinkedIn pages using the GPT_SCRAPE function does not work, so I couldn\u2019t \u00a0just scrape the entire LinkedIn profile then search within it.<\/em><\/p>\n Here are the results:<\/p>\n In this test it got 4 things right:<\/p>\n And got 2 things wrong:<\/p>\n I\u2019m torn here.<\/p>\n On one hand, GPT for Sheets got a decent amount of the prospecting info correct<\/strong>\u00a0and would save me time.<\/p>\n But on the other hand, it was inaccurate enough that I don\u2019t know if I\u2019d trust these results at scale.<\/p>\n Your mileage will vary<\/strong>; it depends on how many prospects you\u2019re researching, what type of information you need, and how damaging it would be to get someone\u2019s info wrong in a message to them.<\/p>\n Once you\u2019ve found and\/or scraped your prospecting info, it\u2019s time to use it in a campaign.<\/p>\n GMass has a native integration with Google Sheets<\/strong>\u00a0so you can go straight from this prospecting work into your campaign without any extra steps.<\/p>\n To begin the campaign process, I connected this Google Sheet to a new GMass campaign. Then I made sure to select the right sheet within my Google Sheets that had the contact info for my prospects.<\/p>\n GMass puts all of the email addresses I found into a campaign<\/strong>. You can also mail merge in anything else you found during the prospecting process.<\/p>\nScraping Prospect Data with GPT for Sheets: Table of Contents<\/h2>\n
\n
\n
How to Use GPT for Sheets for Prospecting<\/h2>\n
Scraping a website\u2019s \u201ccontact us\u201d page, then finding the email address<\/h3>\n
![\"Adding](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image1-3.png\")
<\/a><\/figure>\n![\"Scraped](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image6-2.png\")
<\/a><\/figure>\n=GPT_EXTRACT(B3, \"email address\")<\/code><\/p>\n![\"Finding](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image4-3.png\")
<\/a><\/figure>\nHunting down contact info just by providing a company name<\/h3>\n
=\"Look up the Contact page for the sock company \"&A2& \" and find a contact email address on that page. Give me the URL of the Contact page as well as the contact email address\"<\/code><\/p>\n=GPT_WEB(B2)<\/code><\/p>\n![\"Using](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image3-3.png\")
<\/a><\/figure>\n=GPT_EXTRACT(C2, \"url\")<\/code><\/p>\n=GPT_EXTRACT(C2, \"email address\")<\/code><\/p>\n![\"Splitting](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image2-3.png\")
<\/a><\/figure>\nUsing GPT VISION to search images and documents<\/h3>\n
=GPT_VISION(prompt, url)<\/code>.<\/p>\n![\"An](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image10.png\")
<\/a><\/figure>\n=GPT_VISION(\"Create a comma-separated list of all emails in this image\", \"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/emailsampleimage1.png\")<\/code><\/p>\n![\"The](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image12.png\")
<\/a><\/figure>\n=TRANSPOSE(SPLIT(A2, \",\"))<\/code> would get that done.<\/p>\n![\"Burying](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image8-1.png\")
<\/a><\/figure>\n![\"Handwritten](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image11.png\")
<\/a><\/figure>\nScraping LinkedIn profiles to find prospect information<\/h3>\n
\n
=GPT_WEB(\"Find me the name of the CEO of the email company GMass\")<\/code><\/li>\n=GPT_WEB(\"Find me the name of the CEO of the email company GMass\")<\/code><\/li>\n=GPT_WEB(\"Find the URL of a public profile or interview transcript with \"&A2)<\/code><\/li>\n=GPT_WEB(\"Find the URL of a public profile or interview transcript with \"&A2)<\/code><\/li>\n=GPT_WEB(\"Find the college where \"&A2& \" attended\")<\/code><\/li>\n=GPT_WEB(\"Find an email address for \"&A2)<\/code><\/li>\n<\/ul>\n![\"How](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image7-2.png\")
<\/a><\/figure>\n\n
\n
Using Your Prospected Data in GMass Campaign<\/h2>\n
![\"Connecting](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image9-1.png\")
<\/a><\/figure>\n![\"GMass](\"https:\/\/www.gmass.co\/blog\/wp-content\/uploads\/2024\/06\/image5-3.png\")
<\/a><\/figure>\n