Pinterest Annotations: How Pinterest Labels Your Pins for Search

When you search Pinterest for "sheet pan dinner", Pinterest does not read every word of every pin on the spot. It relies on labels it has already attached to each pin. Pinterest calls these labels annotations, and they are one of the main ways Pinterest understands what a pin is about. Pinterest annotations explain a lot about how Pinterest categorizes pins, and why some pins show up for a search while others don't.

Most of what is public about annotations comes from a 2019 post on the Pinterest Engineering blog. Pinterest's systems have likely changed since then, so read the details below as how Pinterest described the system, not as a guarantee of how it works today.

What Pinterest annotations are

An annotation is a short keyword or phrase, one to six words long, that describes the subject of a pin. Each annotation has three parts: the language, the phrase itself and a confidence score. A pin about a sloth sanctuary in Costa Rica might carry annotations such as "sloth sanctuary", "sloths" and "costa rica", each with a score close to 1.

Annotations are not free text. Pinterest picks them from a curated vocabulary it calls the dictionary, which held around 100,000 terms per language when the engineering post was written. Misspellings, stopwords, fragments and vague words are kept out of it. Pinterest said it extracts annotations in 28 languages.

Where Pinterest gets the words for annotations

Pinterest's engineers listed these text sources for a pin's annotations:

• The pin's title, description and link URL
• The name and description of the board the pin is saved to
• The title and description of the page the pin links to
• Search queries that often lead people to click the pin
• Names of objects that Pinterest's visual model detects in the image

That list matters. Your pin title and description are only part of the picture. The board you save to and the page you link to also feed the labels.

How Pinterest chooses the annotations for a pin

Pinterest's engineering team described the process in a few steps:

1. Detect the language of each piece of text.
2. Break the text into every phrase of one to six words.
3. Clean up each phrase (accents, punctuation, word endings) and keep only the ones found in the dictionary.
4. Score each remaining phrase with a machine learning model, then drop the low scores.

The model looks at how often a phrase appears in the pin's text, how similar the phrase is to the pin, which source the text came from, and how often people search for the phrase. A phrase that people really search, used clearly in your text, has a better chance of becoming a strong annotation.

Where Pinterest uses annotations

According to the same engineering post, annotations feed many parts of Pinterest:

For search, Pinterest stores annotations rather than every word of every pin. Its engineers wrote that annotation scores track relevance better than simple word counts, and that storing only annotations saves a large amount of space across more than 200 billion pins.

How fast new pins get annotations

Pinterest computed annotations for all pins in a weekly batch job, which could lag by several days for brand-new pins. To fill that gap, an "Instant Annotator" service gave new pins annotations within seconds of creation. So a new pin is labelled almost at once, and its labels can be refined later.

Annotations and interests are not the same thing

Pinterest also maps pins to interests: topics in a hierarchy, from broad areas like "DIY and Crafts" down to narrow ones. That system, which Pinterest calls Pin2Interest, uses annotations together with image signals and board names. Some tools call these labels "annotated interests". Annotations describe a pin in specific phrases; interests place it in Pinterest's topic tree.

How to write pins that earn the right Pinterest annotations

You cannot type annotations into a pin. Pinterest generates them. What you control is every source they come from:

1. Name the subject in the title with the phrase people search, near the start. These are your Pinterest keywords.
2. Use the description for related phrases that genuinely describe the pin, written as sentences.
3. Save to a board with a clear, specific name, such as "Sheet Pan Dinners" instead of "Yum".
4. Make the linked page match. The page title and meta description of your blog post are part of the input, so on-page SEO helps on Pinterest too.
5. Show the subject clearly in the image, because Pinterest's visual model contributes labels.

Our free Pinterest character counter checks that your keyword appears early in the title and in the description.

Can you see a pin's annotations?

The Pinterest app does not display annotations. Some third-party Pinterest tools show the labels attached to pins, and PinRankr is designed to show them for the top pins of any search, so you can compare your labels with the ones that rank. See the feature plan or join early access.