I pulled 2,140 real searches about what to post. The popular ones were useless to me.

I wanted to write about what founders search before they post. I did not want to pay for a keyword tool to find out.

Google's autocomplete is free, has no API key, and shows you what people actually type. Here's the method, the 2,140 results, and the finding that changed what I'm going to write next.

The method

The endpoint is public:

https://suggestqueries.google.com/complete/search?client=firefox&hl=en&gl=us&q=YOUR+QUERY

It returns a JSON array of what Google would suggest, ordered by popularity. That ordering is the signal — position 0 is what most people go on to search.

The trick that turns it from a toy into a dataset is alphabet expansion. You don't just ask for build in public — you ask for build in public a, build in public b, all the way to z. Each letter surfaces a different slice of the long tail.

const seeds = ["build in public", "what to post on", "how to grow a saas"];
const letters = "abcdefghijklmnopqrstuvwxyz".split("");

for (const seed of seeds) {
  await suggest(seed);
  for (const l of letters) await suggest(`${seed} ${l}`);
}

Fifteen seeds × 27 queries each = 405 requests, a couple of minutes, and 2,140 unique suggestions.

The clusters

My first reaction was: great, write about engagement rate. 408 variations is enormous, and I already had a page on it.

Then I read the actual queries.

The finding

Here's a real slice of what content ideas for expands into:

content ideas for nail techs
content ideas for estheticians
content ideas for hairstylists
content ideas for jewelry business
content ideas for realtors
content ideas for music artists

Not one of those is a founder. The huge clusters in my market — engagement rate, content ideas, best time to post on Instagram — belong to social media managers and beauty, fitness and real-estate creators.

They're a real audience. They're not mine.

And the queries that are mine are tiny:

how to get first saas users
how to grow b2b saas
reddit marketing for b2b saas
how to get your first customers y combinator

Twenty-two variations. Against four hundred and eight.

Why I'm going after the twenty-two

This is the trade-off nobody says out loud when they tell you to "target high-volume keywords":

Ranking for a popular query in your market can bring you an audience that will never buy from you. You get traffic, your analytics look healthy, and your signup rate quietly tells you the truth six months later.

A nail technician searching "engagement rate calculator" is not going to buy a tool that helps founders grow a SaaS. Neither of us did anything wrong; we just aren't for each other.

Meanwhile how to get first saas users has low volume and near-perfect intent. Whoever types it is, almost by definition, the person I'm building for.

So: intent over volume. A hundred right visitors beat ten thousand wrong ones, and the difference shows up in the only number that matters this early — how many leave an email.

The one that stopped me

Buried in the 883 queries that matched my audience:

how to find trending topics in my niche

That's not adjacent to what I'm building. That is what I'm building, typed by the person who needs it.

I'd been describing the product in my own words. Someone had already written a better description, and Google had been collecting it the whole time.

What this doesn't tell you

Autocomplete gives you ordering, not volume. I know engagement rate calculator is more searched than how to get first saas users. I do not know if that's ten times more or ten thousand times more, and the difference would change some decisions.

For that you need a keyword tool with real numbers, and I'd rather spend the first hundred dollars elsewhere until I know the shape of the thing.

It also only reflects what Google is willing to suggest. Queries that are new, rare, or filtered simply don't appear — so absence here is not evidence of absence in the world.

Run it on your own market

async function suggest(q) {
  const url = `https://suggestqueries.google.com/complete/search`
            + `?client=firefox&hl=en&gl=us&q=${encodeURIComponent(q)}`;
  const [, suggestions] = await (await fetch(url)).json();
  return suggestions;
}

Pick five seeds a customer would type. Expand each with a–z. Then read the results looking for who is asking, not how many.

That second part is the whole exercise. The counts are easy. Noticing that your biggest cluster belongs to someone else's customer is the part that saves you six months.

I'm building HonestHook in public — it tracks what's rising and dying in your niche over time, so you know what to post before you write it. Yes, including the part where I find out what people are searching for.