Which keywords should you advertise on? Turn guesswork into a ranked keyword list

What this is

One of our published methods, written in plain language: what it can do, when to use it, what goes in and what comes out.

Why it matters

Most methods are published as papers or code. This one is written so you can decide whether it fits your problem before you invest time in it.

How to use it

Read the page top to bottom. If it matches your question, use the online tool - or subscribe and run it on your own data.

What you get

A clear answer on what the method does and does not do, plus a live curve from our own demand series. Free to read.

Free to readOnline tools from $5 per runSee it →

Demand curve read by this method — A live reading from our own demand series - the shape of the signal this method works on. 570 industries are tracked daily; this is one of them.

Data base: InfoGap own records - 391 Chinese cities x 8 dimensions, 213,208 city-industry rows, 40,499 POIs (central / lifechain, 2026-10-08).

InfoGap Algorithm Handbook · one algorithm per page, plain language

1In one line: what it does

It turns "which words should I be talking to" from guesswork into a keyword list ranked by priority.

2When you would use it

Three situations you are most likely in right now:

You run search ads and pick the big words that sound hottest, and your cost per click climbs and climbs.

What people usually do: keep buying the big words and switch on automated bidding to save trouble.

Using it: put the industry and the candidate words in, and each word is marked with how many platforms it can be found on and whether anyone really searches it; the ones nobody searches are crossed out for you.

You want to know whether "coffee shop startup" is a word worth writing about.

What people usually do: write on feel, because the word sounds right.

Using it: not found on several platforms → in this version it sits in the lowest band and goes on the elimination list, leaving your effort for words like "coffee shop pain points" that have people on several platforms.

The same topic has people asking on one platform and people posting about it on another, and you do not know which to do first.

What people usually do: do both, splitting your energy in half.

Using it: the output carries which platforms each word lands on and how many results on each, so you start where the people are.

3What goes in · what comes out

Input: industry + candidate keywords + the search results each keyword gets on each platform (below is the set from the real run)

Industry — coffee shop
Candidate keywords — coffee shop pain points, how to run a coffee shop, coffee shop pitfalls to avoid, coffee shop startup
coffee shop pain points — Zhihu 1200 results (trust 0.8), Xiaohongshu 3400 (0.7), Baidu 900 (0.6)
how to run a coffee shop — Zhihu 800 (0.7), Baidu 700 (0.6)
coffee shop pitfalls to avoid, coffee shop startup — no data supplied

What comes out

Output: the run's own text (not a character changed)

coffee shop pain points   → verified on all 3 platforms, score 0.18, priority P3
how to run a coffee shop  → verified on 2 platforms, score 0.08, priority P3
coffee shop pitfalls      → 0 platforms, score 0.0, priority P3
coffee shop startup       → 0 platforms, score 0.0, priority P3

Elimination list: coffee shop startup
Platform list (8):
  Open, searchable from outside the app: Zhihu, Baidu
  Semi-open: Toutiao, Bilibili, WeChat Search
  Closed, only searchable inside the app: Xiaohongshu, Douyin, Kuaishou

How to read it: put the same batch of words together and you see at a glance which have real people behind them and which are self-entertainment. Words like "pitfalls" and "startup" that sound especially right are simply wasted words if nobody searches them. Note that P3 is the lowest band — this input did not verify the words on enough platforms, so they all stick at P3; verify on more platforms and the band moves up.

📈One chart: what interest in a candidate keyword looks like day by day

A keyword is not the same from one day to the next; this line is the evidence for "check it daily, do not guess".

4How to use it

There is no public calculator for this yet — the method above is the whole procedure; on this site the inputs have to be keyed in by hand.

Write down the industry and the candidate keywords
Record how many results each platform returned for each keyword, and whether you judge that source trustworthy
Read the ranking and the elimination list, and do the ones people actually search first

5Who it is for

People about to buy ads who need to pick their words first; shop owners whose content nobody reads; people running accounts for somebody else.

6Common questions

Q: Why look at platforms separately?

A: The same word has different numbers of people in different places. Xiaohongshu and Douyin, for instance, can only be searched from inside the app — you cannot search them from outside — so they have to be counted separately.

Q: Why is everything in the lowest band, P3?

A: Because this input did not verify the word on enough platforms, so the score cannot climb. Only verification across more platforms moves it up.

Q: Can I still use an eliminated word?

A: You can, but do not expect it to bring you people. Spend your effort on words somebody searches first.

Q: Does it go and search by itself?

A: Not in this version. The search results have to be filled in by you — better to say that up front than to have you think it is automatic.

7Related algorithms

funnel-evaluate · where traffic gets stuck demand-verify · is the demand real industry-vitality · is the industry itself any good ← Back to the Algorithm Handbook Handbook index

✓Why this scenario exists (the evidence)

Why this scenario exists:

The most expensive lesson in the overseas scouting report: broad words plus automated bidding, 3 clicks burning 388 dollars, one of those clicks costing 335.80 dollars; another competitor keyword's bill was 8 months, 187 clicks, 4 conversions, 0 customers, at 12.80 dollars a click.

The same report put "search broad match / broad keywords + automated bidding" straight into its top two of things that "look cheap but are actually traps".

Source: the site's overseas ad-scout report dated 2026-09-21.