Research & briefs
Using Google Autocomplete for keyword research, without a paid tool
Autocomplete shows real query phrasing, not search volume. Used for what it measures, it is the fastest free source of subtopics and long-tail variants.
6 min read
Autocomplete is the dropdown that appears while you type into Google. Its predictions come from real searches — Google documents them as reflecting common and trending queries related to what you are typing, filtered by language and location.
That makes it a phrasing instrument, not a volume instrument. It will tell you that people search for "seo brief template free" and "seo brief template notion" but never how many. Treating it as a volume proxy is the single most common way to misuse it.
This article covers the method that produces usable output in about fifteen minutes per topic, what the suggestions can and cannot tell you, the biases Google documents in the system, and how the harvested phrasing feeds a brief.
What Autocomplete actually measures
Predictions are generated from queries that real people search for, weighted towards what is common in your language and region and towards what is trending. They are predictions of what you might be typing, not recommendations and not a ranked list of the most valuable terms.
Google also applies removal policies. Predictions that are violent, sexually explicit, hateful, dangerous, or that name individuals in potentially defamatory ways can be removed, as can predictions relating to elections. Some categories of query therefore show a much thinner dropdown than the underlying search behaviour would suggest.
| Signal | Reliable conclusion | Unsupported conclusion |
|---|---|---|
| A suggestion appears | People phrase the query this way in your region and language | The query has meaningful monthly volume |
| A suggestion appears high in the list | It is comparatively common or trending for that prefix | It has more volume than a suggestion listed below it |
| A suggestion never appears | It is uncommon for that prefix, or filtered by policy | Nobody searches for it |
| Suggestions differ between two machines | Personalisation, location or language differ | The data changed |
The method: prefix, modifier, alphabet
Three passes, in order. Each one surfaces a different class of query, and the third is only worth running once the first two stop producing new phrasing.
- Pass one — the bare seed — Type your topic and nothing else, then read the ten suggestions. These are the dominant framings. If none of them matches the page you planned, that is a finding: your topic is phrased in a way the market does not use.
- Pass two — question and comparison modifiers — Prefix the seed with how, what, why, when, is, does, can, best, vs, alternative, template, example, free. Each prefix returns a different dropdown, and questions and comparisons map directly to subheadings.
- Pass three — the alphabet sweep — Append a space and each letter a through z. This forces the system to complete in twenty-six directions and reliably surfaces long-tail variants the first two passes miss. Stop when three consecutive letters return nothing new.
- Pass four — the trailing wildcard — Type your seed, then a space and an underscore or a word you expect mid-phrase. Google completes around what you typed, which surfaces phrasings where your seed sits in the middle rather than at the start.
Reading the harvest
Group by modifier, not alphabetically
Sort what you collected into questions, comparisons, qualifiers (free, best, simple) and format words (template, example, checklist, tool). The grouping is the analysis: a topic dominated by format words needs a downloadable or fillable artefact, while one dominated by questions needs an answer-first structure with each question as its own passage.
Separate subtopics from separate pages
A suggestion is a subtopic when the answer belongs inside the page you planned, and a separate page when it has a different intent. "seo content brief example" is a section. "content brief vs creative brief" is its own page with a comparison format, and forcing it into the first one weakens both.
The reliable way to decide is to search both queries and compare the results pages. If they return substantially the same URLs, one page can serve both; if they diverge, you need two.
Cross-check against People Also Ask
Autocomplete gives you the phrasing people type; People Also Ask gives you the questions Google considers adjacent to the query it resolved. They overlap, but the gaps in each are useful — a PAA question absent from Autocomplete is often a subtopic your competitors have not written for.
Where the method misleads
The last row is the expensive one. Autocomplete makes it trivially easy to generate a list of forty near-identical long-tail phrases, and building a page for each is how sites end up cannibalising themselves. Variants are phrasing evidence for one page far more often than they are page ideas.
| Trap | What goes wrong | How to avoid it |
|---|---|---|
| Treating order as volume | You prioritise a suggestion that is merely trending this week | Use the list for phrasing; confirm demand with a volume source before committing budget |
| Researching in your own session | Personalisation returns your own history back to you | Incognito window, explicit region and language |
| Brand names in the dropdown | You target a query where a competitor is the intent | Search it — navigational queries for another brand are not winnable |
| Assuming a gap means opportunity | A missing suggestion may be a policy removal, not an unserved need | Check whether the topic falls in a filtered category before concluding |
| Harvesting hundreds of variants | You produce a spreadsheet, not a decision | Stop when new letters return rephrasings of what you already have |
| One page per variant | Near-duplicate pages competing with each other | Merge variants that return the same SERP into one page |
Turning the harvest into a brief
The output of this exercise maps onto exactly three fields of a content brief. Everything else in the brief comes from elsewhere.
- Primary query — the dominant framing from pass one, confirmed against a live search rather than chosen for its wording.
- Must-cover subtopics — the question and qualifier groups, filtered to those that belong on this page.
- Section phrasing — the exact words people use, which is what keeps headings recognisable instead of abstract.
Free sources that complement it
Search Console deserves emphasis: for an existing page, its query report tells you which phrasings already bring impressions. Queries with impressions and a poor position are the strongest candidates for a rewrite, because demand is proven and the gap is yours.
- People Also Ask on the live SERP — adjacent questions, expandable in place, and a direct source of subheadings.
- Related searches at the foot of the results page — broader reformulations rather than completions of what you typed.
- Google Trends — relative interest over time and by region, which is the closest free substitute for the volume Autocomplete does not give you.
- Search Console query data for pages you already have — the only source here that reports what people searched before landing on your site specifically.
- The Autocomplete dropdowns on YouTube and other large search surfaces, when the topic has a video or marketplace dimension.
Questions people ask
Does Google Autocomplete show search volume?
- No. Predictions reflect common and trending queries for what you are typing, but Google publishes no volume figures with them and the order is not a volume ranking. Use Trends or a keyword tool when you need numbers.
Why do I get different suggestions than a colleague?
- Predictions vary by language, location and recent activity. Research in a private window and fix the region you are targeting, otherwise you are comparing two different datasets.
Is scraping Autocomplete allowed?
- Automated querying of Google Search, including the suggestion endpoint, is restricted by their terms. Manual research and licensed third-party tools are the safe route.
How many suggestions should I collect before stopping?
- Stop when three consecutive letters in the alphabet sweep return only rephrasings of what you already have. Past that point you are collecting variants of one page, not finding new pages.
References
- Google Search Help — How Google Autocomplete predictions work — Official explanation that predictions reflect real, common and trending searches, and vary by language and location.
- Google Search Central Blog — How Google Autocomplete works in Search — Google on why predictions are not recommendations and how the removal policies are applied.
- Google Search Console — Performance report — The query-level impression and position data used to validate phrasing on pages you already have.
- Google Trends — Relative interest over time and by region — the free way to sanity-check demand for a harvested phrase.
Keep reading
- Confirming a variant deserves its own page — Comparing two results pages to decide between one page and two.
- What the modifiers tell you about intent — Why "best" and "how to" are hints rather than verdicts, and how to check them.
- Where the harvested phrasing lands — The three brief fields this research fills, and the six it does not.
- Turning collected questions into passages — How to structure each harvested question so the answer stands on its own.