How to Choose Search Phrases

The short version

A search phrase is the question someone asks an AI engine before they know your business exists. indextr asks each engine your phrases on a schedule and records whether you get named, how prominently, and who gets named instead of you.

That means your phrases are not a description of your business. They are a guess about the words a stranger uses when they have a problem and no idea who to call.

A common first mistake

Businesses often describe themselves in the language of their industry. Customers describe what they are looking for in their own words. When the two differ, a phrase written in your language measures a search a customer may never make.

If a phrase reads like something you would put on your own website, it is probably the wrong phrase. Read it back and ask whether a person who has never heard of you would type those words.

Never use your own name

indextr will not suggest a phrase containing your business name, and you should not add one. Asking an engine about your business by name tests whether it can describe you once it has been told who you are. It does not test whether anybody finds you.

The phrase has already handed the engine the answer, so the result says nothing about being found. The customer you are worried about is the one who has never typed your name in their life.

A phrase has to be able to name somebody

Before you add a phrase, ask: could the answer to this name a specific business? If not, the phrase cannot help you — no matter how relevant it looks.

Cost, advice and how-to questions are the ones to check hardest against this test. They can make good blog topics; a phrase whose answer names nobody is no use as a measurement.

If you are local, name your city

A phrase for a local business should name its city. Not because it reads better, but because the engine has to put you somewhere — and if you do not say where, it chooses for you.

When we ran business coach near me with no location attached, the results came back for Venice and Fort Myers, Florida. Nobody asked for Florida. Left to itself, the search simply landed somewhere, and everything measured after that was measured about the wrong place.

A national or online business is the opposite case: it has no reason to use a city or near me, because it has no near. Its phrases name the category without a place.

The kind worth tracking

Problem phrases (“my parents need a will and live in another state”) and comparison phrases (“estate planning attorney vs online will service”) feel as if they should count. On the account where we measured them, they failed the test above: the engine gave advice or an explanation and named nobody. When we measured problem phrases on 1 September 2026, they named a business on 0 of 3 runs on four engines and 1 of 3 on ChatGPT, against 3 of 3 for a category phrase on the same account. A comparison phrase we measured returned no businesses at all on three engines.

How many to track

Fewer than you might think. indextr includes 15 phrases on Starter and 30 on Pro.

The list is not a net to catch everything a customer might conceivably say. It is the set of questions you would genuinely change your website to win. We think twelve phrases you would act on beat thirty you would not — because every phrase counts toward your score whether or not you ever intend to do anything about it.

Changing your phrases later

You should. The first list is a guess, and the first few scans tell you which guesses were wrong. Phrases that never name anybody are the first to go.

One thing to know before you do: your score is calculated across whichever phrases were active at the time of each scan, so changing the list changes the measurement as well as the result. indextr records which phrases every scan covered and marks the change on your trend chart, so that a step in the line is labelled as your own edit — rather than being mistaken, three months later, for something that happened in the market.

What about tools that generate a page for every phrase?

There are tools that will produce hundreds of pages, each written to answer one specific search. The pitch is straightforward: answer every question and you will be found for all of them.

Two of Google's spam policies apply, and the wording matters. Scaled content abuse is “when many pages are generated for the primary purpose of manipulating search rankings and not helping users” — with “using generative AI tools… to generate many pages without adding value” given as an example. Doorway abuse is “when sites or pages are created to rank for specific, similar search queries”, and one of Google's own examples is pages targeting particular cities that funnel everyone to the same destination.

Google Search Central — spam policies · checked 29 August 2026

Note what the test actually is: value, not volume. Generated pages are not banned. A site with genuinely different information on each page — real listings, real inventory, real data — is doing something useful at scale, and that is fine. The problem is pages that differ only by a swapped town name and funnel to the same contact form.

Which is where a local business runs out of road. Once the pages stop carrying genuinely different information, you are producing the pattern the policy describes, whether a tool wrote them or you did.

But the more useful point is that those tools sit downstream of this page. Whatever you use to write the answers, you still have to know which questions are worth answering, and whether answering them changed anything. Knowing which questions matter is what makes the work finite.

A short checklist

indextr proposes a starting set from your website when you sign up, and suggests new ones later based on what the engines actually said in your own scans. You can edit any of them at any time — the suggestions are a starting point, not the answer.