Google Search vs AI Assistants: Where Buyers Actually Ask the Question

Search is dying, and Google is bigger than ever

Two headlines land in your feed on the same morning. One says search is dying, because everyone asks AI now. The other says Google is bigger than ever.

Both come with charts. Both charts are real. So which one do you plan your content around?

Neither. The split between Google Search and AI assistants does not depend on the year. It depends on the topic, and you can measure it.

We measured it on four topics we write about. Three were mostly searched. One was asked of AI assistants more than three times as often as it was searched.

That one changed what we build. This guide gives you the same measurement and the page each result calls for.

Is AI search replacing Google search?

Not by volume. Google still handles far more searches than all AI assistants combined, and the current counts sit in the dated box near the end.

But Google itself now answers many questions on the results page. A summary appears above the links. Fewer people click through.

What decides whether that summary shows up? Mostly the shape of the question.

Pew Research Center tracked the browsing of 900 adults. One- and two-word searches rarely produced an AI summary. Searches of ten words or more produced one about half the time.

So the real split is not Google against AI. It is short phrases against long questions.

A short phrase ends in a list of links. A long question ends in an answer, inside Google or inside an assistant.

Good news, then. You do not need to guess where your buyers went. You need to know which kind of question your topic is.

The Ask-Ratio: one number per topic shows where buyers ask

The Ask-Ratio is a topic's monthly AI assistant asks divided by its monthly Google searches. Asks are the questions people type into an AI assistant about that topic. Some keyword tools now estimate both numbers.

Why a ratio and not two totals? The authors of Lean Analytics make the case that the best metrics are ratios. A ratio compares by design, and you can act on it.

Their test for any metric is simple. If it moves, what will you do differently?

Two totals tell you size. The ratio tells you which page to build.

They also say to draw your line in the sand before the data comes in. So set the bands first, then pull the numbers.

Under 0.2: a search topic. For every ten searches, fewer than two people ask. Win the list of results.

0.2 to 1.0: a mixed topic. The same buyer searches the short version and asks the long one. One page has to rank and be quotable.

Over 1.0: an asked topic. More people ask than search. Win the answer.

The lines are ours. They are a working rule, not a law. At 1.0 both places are even, and below 0.2 the asking crowd is too small to plan a page around.

We measured four of our topics, and one flipped

Here is the audit on four topics we cover. The numbers are monthly US estimates from one keyword data provider.

"AI consultant" looks like a search topic. 8,100 Google searches a month against 717 AI asks. Ratio 0.09.

People want a list of firms to compare. A results page is a list.

"AI transformation" and "AI compliance" are mixed. Ratios of 0.32 and 0.44. That is about one ask for every two or three searches.

"AI operations" flipped. 320 Google searches a month against 1,079 AI asks. Ratio 3.4.

More than three people asked for every one who searched.

Same industry. Same buyers. Opposite answer.

Four topics are not a study. They are enough to kill the idea that one answer fits your whole business.

Read the ratio as a range, because the tools disagree

Here is the catch the headlines skip. Keyword tools estimate. They do not count.

We checked "AI consultant" in a second tool. It showed 3,300 Google searches a month, not 8,100.

Same phrase. Two tools. One estimate is more than double the other.

So the Ask-Ratio is 0.09 with one tool and 0.22 with the other.

That range crosses our 0.2 line. Is the measurement useless, then? No.

In How to Measure Anything in Cybersecurity Risk, Douglas Hubbard and Richard Seiersen define measurement as any observation that cuts your uncertainty. It never has to be exact to be worth taking.

Before the check, you had no idea where buyers ask about this topic. After it, you know it sits between 0.09 and 0.22. That is enough to decide.

Our rule: if the whole range sits inside one band, build for that band. If it crosses a line, build the mixed page, because it serves both sides.

How many readings do you need? Fewer than you think. The book's rule of five says that in a random sample of five, the true middle value lands between your lowest and highest reading about 94 percent of the time.

Months are not a perfect random draw, since some topics have seasons. So treat it as a rough guide. Pull five months from two tools if you can, then keep the lowest and highest ratio.

Why some topics get asked instead of searched

Our explanation starts with where the buyer is on the road to a purchase. Bob Moesta's Demand-Side Sales 101 maps that road in stages, and two of them matter here.

Passive looking is the stage where a person feels the problem but has no plan. They are learning the problem and the words for it.

Active looking starts once they set aside time or money to fix it. Now they gather facts and compare options.

The next part is our reading, not the book's claim. Passive lookers describe. Active lookers compare.

Someone who describes writes a paragraph about their own company. That is an ask.

Someone who compares types two words and wants a list. That is a search.

It fits our four topics. "AI operations" is a mess you describe. "AI consultant" is a shortlist you compare.

So a high ratio may also mean your buyers are early and still learning. Check that guess against your own sales calls before you bet a quarter on it.

Build the page each band calls for

Each band gets one page type. There are no extra technical requirements for Google's AI features, and the usual SEO basics still apply. So the work is what goes on the page, not a secret switch.

Search topic: put the exact phrase in the title. Show the options side by side. Then earn the links that move you up, which is the subject of the Page-2 Problem.

Mixed topic: open with a short, direct answer. Then let the headings carry the questions buyers ask. One page, two jobs.

Asked topic: answer in the first two sentences. Use your own numbers. Make every section make sense alone, because an assistant may lift one section and skip the rest.

Google's own advice on helpful, people-first content points the same way. Write for the person asking, not for the engine.

Which sources do assistants actually quote? We counted them in what AI answer engines actually cite. To see whether they name you, use the AI search visibility system.

A page built for the wrong band can still be a good page. It just answers a question nobody there is asking.

Picking a side is the most expensive way to get this wrong

How to waste a year of content in four easy steps:

1) Read one headline about AI search.

2) Rewrite every page for AI assistants, including the ones people only ever search.

3) Never measure a single topic.

4) When traffic drops, blame the algorithm.

That is not a strategy. That is a mood swing with a budget.

Our take: the loudest voices in this debate usually sell one side of it. Search shops say search is fine, and AI visibility vendors say search is dying.

Both pitches skip the same step. Neither one looked at your topics. That skipped step is the bullshit you end up paying for.

Four places the audit goes wrong, and the fix

Your tool shows zero AI asks. Zero in one tool is not zero demand. Many tools have no AI volume at all, and small topics fall below their floor.

The fix is other sources. You have more data than you think: the questions on your sales calls, in your contact forms and in your site search. Our customer truth system shows how to collect them.

You measured a phrase, not a topic. "AI ops" and "AI operations" are one topic. Add the variants together before you divide.

You trusted one month. A single month can catch a spike or a slump. Use five, and read the range.

The ratio moved and nobody noticed. Chip Huyen points out in AI Engineering that users learn to phrase queries to get better results, the way they did with Google. Phrasing shifts, and the ratio shifts with it.

So a ratio is a reading, not a fact. It is a weather report, not a birth certificate.

Recheck twice a year, because the places people ask can move fast

The places people ask can change faster than most content plans. Principles of Marketing Engineering and Analytics, by Gary Lilien, Arvind Rangaswamy and Arnaud De Bruyn, shows that Facebook was missing from the top ten US web brands at the end of 2007.

By 2016 it ranked third. Nine years took it from absent to near the top.

Will AI assistants move that fast? Nobody knows. That is exactly why you measure instead of guess.

Our rule is a recheck every six months. Same topics, same tools, same bands.

When a topic changes band, change its page. When it does not, leave the page alone and spend the time elsewhere.

The numbers that change with every study sit in the box below, with their dates. The method above does not depend on any of them.

Run the Ask-Ratio audit with AI this week

The prompt below runs the whole audit in your preferred AI. It asks for your topics and your numbers and never invents one. You get a range, a band and a page type for each topic.

This week, do three things. List your ten main topics. Pull five months of Google searches and AI asks for each, from two tools if you can.

Run the prompt. Then fix one page first: the one that is the worst match for its topic's band.

Put a recheck in your calendar six months out. If you want a second pair of eyes, our SEO team runs this audit as part of every content plan, and hub and spoke SEO shows how to group the topics first.

Remember the two headlines? Both were right. Each was right about a different kind of topic.

So stop arguing about which road your buyers take. Count the cars on each one.