Turn Google Autocomplete into a buyer-intent keyword map

WHY THIS WORKFLOW

Use autocomplete for phrasing, not imaginary precision

Keyword research often mixes two different jobs. The first is discovery: finding the words, questions, comparisons, and qualifiers people encounter while searching. The second is validation: estimating volume, cost per click, difficulty, or conversion value. Google Autocomplete is useful for the first job. It is not a substitute for the second.

That distinction matters because an autocomplete suggestion can reveal a useful angle even when you do not yet know its monthly volume. Phrases containing terms such as best, vs, price, reviews, alternative, near me, or for a specific audience can point to decisions a searcher is trying to make. Your job is to collect those phrases consistently, preserve their locale and source seed, then decide which deserve further validation.

STEP 1

Choose seeds close to a real decision

A broad seed such as “electric car” is useful for exploration, but it mixes informational, navigational, and commercial ideas. Add a second pass with phrases that are closer to a decision. The goal is not to force every result to be transactional; it is to make the intent visible enough that you can sort it later.

Keep different products, audiences, and countries in clearly named batches. When every idea goes into one giant run, the output may be larger but the reasoning becomes weaker. Small batches make it easier to see which seed produced a suggestion and to repeat the research later.

STEP 2

Start bounded, then expand deliberately

The safest first run uses a small seed list, one market, and a hard maximum. Leave alphabet and question expansion off until you have confirmed that the language, country, and output fields are correct. This prevents a discovery test from silently becoming a much larger collection job.

{
  "queries": [
    "best electric car charger",
    "electric car charger vs",
    "electric car charger price",
    "electric car charger for"
  ],
  "language": "en",
  "country": "us",
  "expandAlphabet": false,
  "includeQuestions": false,
  "maxResults": 50
}

After inspecting the first dataset, enable A–Z expansion when you need broader long-tail coverage. Enable question expansion when the next deliverable is an FAQ, support library, comparison guide, or educational content map. Increase maxResults last, not first.

STEP 3

Read each row as evidence, not just a keyword

A useful export should retain more than the final phrase. The Actor keeps the original query, the expanded seed that triggered the request, suggestion rank, language, country, source, and collection time. Those fields let you trace why a phrase appeared and make later comparisons less ambiguous.

STEP 4

Cluster by the decision behind the phrase

Do not rank a phrase as “high intent” just because it contains one commercial word. Read the complete phrase and group it by the decision it appears to support. A practical first-pass taxonomy is enough; you can refine it when the patterns in your own dataset become clear.

Deduplicate exact phrases, but do not collapse genuinely different modifiers too early. “Best charger for apartments” and “charger installation cost for apartments” may share a topic while supporting different decisions. Preserve both until the content or product team chooses the final page structure.

STEP 5

Validate before you invest

Autocomplete gives you candidates. Before building an expensive landing page or content cluster, add the signals your decision actually needs: search volume, CPC, ranking difficulty, current SERP type, product margin, conversion history, or qualitative customer evidence. A low-volume term may still be valuable for a high-ticket product; a frequent term may be irrelevant to what you sell.

A simple prioritization sheet can combine four columns: relevance to the offer, closeness to a buying decision, evidence from an external volume source, and the strength of the page you can realistically create. Keep the raw autocomplete export unchanged and score a separate working copy.

AUTOMATION

Call the same bounded workflow through the API

Once the input is producing useful clusters, save it as an Apify task for scheduled research or call the Actor from an application. The same output can be exported to JSON, CSV, or Excel, connected to a webhook, or made available to an AI workflow through Apify MCP.

const response = await fetch(
  "https://api.apify.com/v2/acts/signal_lab~google-autocomplete-keywords/runs?token=APIFY_TOKEN",
  {
    method: "POST",
    headers: { "content-type": "application/json" },
    body: JSON.stringify({
      queries: ["best electric car charger", "electric car charger vs"],
      language: "en",
      country: "us",
      expandAlphabet: false,
      includeQuestions: false,
      maxResults: 50
    })
  }
);

const run = await response.json();

Keep credentials in environment variables rather than source code. For recurring research, keep the country, language, seed set, and result cap fixed so changes in the dataset are easier to interpret.

TROUBLESHOOTING

Common reasons a dataset feels wrong