AI WORLD HQSpot the mood behind customer reviews with Hugging Face's free tool
Paste a review into a free web tool and instantly see if it reads as positive or negative — no setup required.
In short: Hugging Face hosts a free web tool that reads a short text and labels it Positive or Negative with a confidence score. Paste one review at a time, copy the label, and you can sort feedback by mood without rereading every word.
Customer reviews pile up faster than any small team can read them. Hugging Face — a popular AI platform — offers a free sentiment analysis tool on its website that does the first pass for you. You paste a review, the tool tells you whether it leans positive or negative, and how sure it is. This guide walks you through it step by step, using only your browser.
💡 Tip: tap a step’s number when you finish it — a green tick appears and your browser remembers how far you got.
- A free Hugging Face account — sign up at huggingface.co with an email. No credit card needed.
- A computer with a web browser (Chrome, Firefox, Edge, or Safari). The site loads on a phone, but a computer is easier for copy-paste.
- A small batch of reviews in a text file — one or two sentences each. Grab them from Google reviews, Amazon, your survey responses.
- About 10 minutes total. Plan to read each result, not just trust it.
- Honest expectation: this is a first-pass filter, not research-grade analysis. It often misreads sarcasm, very short comments ("ok"), or reviews that mix feelings in one sentence. You'll review the low-confidence results yourself.
Find the sentiment tool on Hugging Face
In your browser's address bar, type huggingface.co and press Enter. Once the page loads, look for the search bar at the top of the screen. Click inside it and type sentiment analysis, then press Enter. A list of AI models appears. Look for one near the top called distilbert-base-uncased-finetuned-sst-2-english (the name looks technical — that's normal). Click on that result.
What you'll see: a model page with a description and an interactive area. Somewhere on that page there should be a text box where you can type or paste text. Look for placeholder text like "Type something…" or "Input text here." Above or beside the box, you'll see a button to make the tool run.
What happens after: the page is ready to accept text. You don't need to read the rest of the page — just stay focused on the input box.
If it looks different: Hugging Face redesigns its site from time to time. If you don't see a search bar at the top, look for a magnifying glass icon 🔍 instead. If the model page doesn't show an input box, scroll down — it's sometimes lower on the page than you'd expect.
You'll know it worked when you can see a clickable text box with a button underneath that looks like it would submit your text.
Sign in so the tool will accept longer text
Click the button on the tool page without typing anything in the box first. A small message usually appears saying something like "You must be logged in" or "Authentication required." That's normal. Look for a Login or Sign in option at the top-right corner of the page. Enter your email and password, or create a free account (it takes about a minute: email, password, confirm).
What you'll see: after signing in, the page reloads and your profile picture or initials appear in the top-right corner where the Login button was.
What happens next: the tool is now ready to process text from your account. Some browsers stay logged in across visits, so you may not see the prompt at all — just move on.
If it looks different: if you can't find Login at the top-right, look inside a menu accessed by your profile icon or by three dots in the corner.
You'll know it worked when your name or avatar shows in the top-right of the page.
Paste one review and run the tool
Open your text file of reviews. Copy just the first review — one or two sentences. Click into the text box on the model page and paste the review. Then click the button underneath the box to submit it.
What you'll see: a short wait (usually 1–3 seconds), then a result panel appears. It shows a label — usually POSITIVE or NEGATIVE in capital letters — followed by a percentage like 98.67%. That percentage is the model's confidence: how sure it is about its answer.
What happens next: the original review stays in the box and the result sits below it. You can now record the result somewhere and move on to the next review.
If it looks different: sometimes the result shows two labels instead of one — for example "POSITIVE: 97%" and "NEGATIVE: 3%". That's normal; it means the model is very sure. Just read the bigger number.
You'll know it worked when you see a percentage above 50% next to either POSITIVE or NEGATIVE. Anything below 80% confidence is a "soft" result — the model isn't sure, so that review needs a human read.
Record the result somewhere you'll find it
This is where the time savings actually show up. Open a free tool — Google Sheets, Excel, or even a plain text file. For each review, write the review text in one column (or one line) and the model's label and confidence in the next column.
What you'll see: a growing list that looks like a table. As you go, you'll start to see patterns — most reviews positive, a clear cluster of negatives, a few mixed-feeling ones with low confidence.
What happens next: when you're done, you can sort the list by the label column. In Google Sheets, click the column header, then look for a sort option (the exact menu name varies — try Data → Sort range, or right-click the column and pick "Sort"). The negative reviews jump to the top automatically. That's your shortlist for follow-up.
If it looks different: if you don't want a spreadsheet, even a plain text file works. Just write the review and the label next to it, like "Broken after a week — NEGATIVE 99%."
You'll know it worked when you can scan your list and instantly see which reviews need a human reply.
Repeat for each review, then spot the soft results
Go back to Step 3 and process the next review: paste it, submit, copy the result into your table, move on. After you've run a batch — say, ten reviews — scroll through your table and look for any result where confidence is below 80%. Those are the ones the model wasn't sure about.
What you'll see: roughly 10–20% of your reviews fall into this "unsure" zone. That's expected.
What happens next: open each soft result and read it yourself. These often contain sarcasm ("Oh sure, great service"), mixed feelings ("Shipping was great but the product is awful"), or very short comments ("meh") that confuse simple models. You decide what to do with each one — positive, negative, or neutral — based on your own judgment. The tool saved you from re-reading the easy 80%; you only focus on the tricky ones.
If it looks different: if nearly every result comes back below 60% confidence, the reviews may be in another language or very technical. The English-only model struggles with both. In that case, the tool isn't useful for this batch — try a different sample.
You'll know it worked when you have a clean list where the obvious complaints are flagged and only the ambiguous handful needs your eyes.
- Trusting a 55% confidence result. Anything below 80% is the model shrugging. Treat those as "unsure" and read them yourself — a coin flip is more accurate than the tool at that confidence.
- Using reviews in another language. This particular model is English-only. Spanish, French, or any other language will return random or low-confidence scores. For multilingual reviews, look for a model labeled "multilingual" in the search results instead.
- Expecting it to catch sarcasm. It usually doesn't. A sarcastic "best purchase ever 🙄" may be tagged POSITIVE because of the word "best." Always skim the low-confidence results for irony.
- Forgetting to log in. Without an account, the widget may refuse to run, or may give you just one free try before asking you to sign up.
Open a new tab, go to huggingface.co, search for "sentiment analysis," click the first result, sign in, and paste this single review to test it end-to-end: "Honestly, this was the worst customer experience I've had in years. Nothing worked and nobody replied to my emails." Click the run button. You should see NEGATIVE with confidence above 95%. That one test takes less than two minutes — and once it works, the rest of the guide will too.
❓ Quick questions
How long does this take?
About 6 minutes — the guide has 5 steps, and you can tick each one off as you go.
Which tool do I need?
This guide uses Hugging Face Hugging Face — but the approach works very similarly in other AI assistants.
Do I need to prepare anything?
- A free Hugging Face account — sign up at huggingface.co with an email. No credit card needed.
- A computer with a web browser (Chrome, Firefox, Edge, or Safari). The site loads on a phone, but a computer is easier for copy-paste.
- A small batch of reviews in a text file — one or two sentences each. Grab them from Google reviews, Amazon, your survey responses.
- About 10 minutes total. Plan to read each result, not just trust it.
- Honest expectation: this is a first-pass filter, not research-grade analysis. It often misreads sarcasm, very short comments ("ok"), or reviews that mix feelings in one sentence. You'll review the low-confidence results yourself.
What mistakes should I avoid?
- Trusting a 55% confidence result. Anything below 80% is the model shrugging. Treat those as "unsure" and read them yourself — a coin flip is more accurate than the tool at that confidence.
- Using reviews in another language. This particular model is English-only. Spanish, French, or any other language will return random or low-confidence scores. For multilingual reviews, look for a model labeled "multilingual" in the search results instead.
- Expecting it to catch sarcasm. It usually doesn't. A sarcastic "best purchase ever 🙄" may be tagged POSITIVE because of the word "best." Always skim the low-confidence results for irony.
- Forgetting to log in. Without an account, the widget may refuse to run, or may give you just one free try before asking you to sign up.
Keep reading
✦ Original step-by-step guide by AI World HQ's AI editorial team. Written in plain language, reviewed for accuracy.
← Back to all stories