ML Models & Algorithms
Predictive Text Models
In shortModels that predict the rest of a word or query as someone types, such as Google and YouTube autocomplete, and a rich source of real query ideas.
Predictive text models suggest how a word or search query is likely to end while someone is still typing. Autocomplete in Google Search, YouTube and Maps, and the word suggestions on a phone keyboard, are all predictive text in action. For SEOs, the suggestions are a direct window into how real people phrase their searches.
How it works
The models learn from large volumes of past input. Google explains that its autocomplete predictions reflect real searches, including common and trending ones, and are shaped by the language and location of the person typing. Predictions that break Google’s policies are removed. Under the hood, predictive text has moved from simple statistical models that count which words follow which (the idea behind n-grams) to neural language models.
You can collect suggestions at scale in several ways: the suggestion endpoints behind Google Search and YouTube, Query Autocomplete and Place Autocomplete in Google’s Places API, and retail autocomplete in Vertex AI Search for commerce.
How it’s used in SEO and marketing
- Long-tail discovery. Append each letter of the alphabet to a seed keyword and collect hundreds of suggestions in one run.
- Questions and modifiers. Reveal the “how”, “why”, “best” and “vs” phrasings people use around a topic.
- Platform differences. Compare Google and YouTube suggestions to decide which topics suit video.
- Local research. Use Places API suggestions for location-based queries.
- Query paths. Suggestions show the refinements people make, which helps map query paths and plan content sequences.
- Early trends. New phrases often appear in autocomplete before keyword tools have volume data for them.
Example
MLforSEO’s autocomplete notebook takes a list of seed keywords such as “home gym”, queries Google Search, YouTube and the Places API, and appends letters a to z to each seed. The output is a de-duplicated table of seed and suggestion pairs that can be clustered and merged into your semantic keyword universe.
Limitations and things to watch
- Suggestions come with no search volume; validate them with keyword tools or Trends data.
- Results vary by location, language and sometimes personal history, so set these parameters explicitly.
- The Google Search and YouTube suggestion endpoints are not officially documented APIs, so they can change or rate-limit without notice. Add pauses between requests.
- Filtered predictions mean sensitive topics may be under-represented.
Learn it on MLforSEO
Where Predictive Text Models shows up on MLforSEO
Learn it in a course
Sources and further reading
- How Google autocomplete predictions work (Google blog)
- How Google autocomplete works in Search (Google Search Help)
- How to use Google Autocomplete API and Places API for Keyword Suggestions with Python MLforSEO
- Multi-Platform Autocomplete Keyword Research with Google Search, YouTube, and Places APIs (Notebook) MLforSEO
Other terms from ML Models & Algorithms
- BERT (Bidirectional Encoder Representations from Transformers)Google's 2018 transformer language model that reads text in both directions to produce context-aware embedding…
- BERTopicA modular topic modelling technique that clusters transformer embeddings and labels each cluster with class-ba…
- BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies)BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies) clusters very large numeric datasets by s…
- BM25A classic ranking function that scores documents by how often query words appear, adjusted for word rarity and…
- Boyer-MooreA fast exact string-search algorithm that compares a pattern from right to left and skips ahead after mismatch…
- c-TF-IDFClass-based TF-IDF: treats each cluster as one document to find the words that best describe it, used by BERTo…
- Centroid-based ClusteringA clustering family that groups data around central points (centroids), each the average of its cluster's memb…
- Convolutional Neural Network (CNN)A deep learning architecture that scans data with small learned filters to detect local patterns; the classic…
- DBSCANDensity-based clustering that groups closely packed points, finds the number of clusters itself and labels iso…