How to Adjust Your Content Strategy Now That Nearly Half of Americans Use AI Chatbots

Step 1: Start From What Changed

How many of your buyers ask AI

Pew surveyed 5,119 US adults in February 2026. Use of AI chatbots rose to 49%, from 33% in 2024 (Pew notes the earlier question was asked differently). ChatGPT alone is used by 44% of US adults, against 34% a year earlier on a question that did not yet list other chatbots, and Gemini by 24%. About a quarter use a chatbot every day.

What happens to the click

Google's own results now carry AI answers too. Pew tracked the browsing of 900 US adults in March 2025: an AI summary appeared on 18% of their Google searches, and when one did, they clicked a traditional result on 8% of visits, against 15% when it did not. They clicked a link inside the summary on 1%.

Put together, the two studies change what a content plan is for. On the questions where an answer appears, being named in that answer is the visibility, and the click comes later if it comes at all. A plan judged only by rankings and sessions measures a shrinking part of what your buyers see.

Step 2: Plan Around the Questions Buyers Ask an Assistant

A keyword list tells you what people type into a search box. An assistant gets a question, often a long one with the buyer's situation in it. Write down the questions a buyer asks before they have a shortlist, in their words, and make them the unit of the plan.

Keep the questions whose answer can name you

Not every question is worth a page. In our tracking, 14.8% of answers named no company at all: definitions, explainers, yes-or-no answers. A brand appeared in 4.3% of those, against 32.5% of answers that named companies. Much of that gap is built in, since an answer that names no company rarely names yours, and that is the point: a question whose answer never names anyone cannot show you a win. Plan around questions whose answer is a shortlist, a comparison or a recommendation. Our guide to choosing prompts to track shows which kinds of question do that most often.

Write them the way buyers talk

"Payroll software" is a keyword. "What payroll software works for a 40-person company with staff in three states" is the question. Keep the situation in it, because the answer changes with it, and so do the pages AI cites.

Ask on the engines your buyers use

Going by Pew's numbers, start with ChatGPT, then Gemini and Google's AI answers. The same question can get different companies and different cited pages on each engine, so treat each engine's answer as its own result.

Step 3: Fund the Formats AI Cites

Most content calendars are full of articles. AI engines cite a narrower mix. Across citations to content pages in our study, ranked lists drew 35.7%, general articles 31.0%, how-to guides 9.7%, comparisons 4.5% and checklists 1.5%.

Shift the calendar toward lists and how-tos

A ranked list answers a shortlist question directly, which is why it is the highest-leverage page in AI search. How-to guides answer the "how do I" questions that come next. Move some of next quarter's article slots to these two formats before you add more slots.

Read the cited pages before you brief

The winning format changes from question to question. For each question you plan a page for, open the pages AI cites and note their format, then write in that format. Our one-question workflow walks through this, from reading the answer to publishing the page.

Step 4: Budget for Pages You Don't Own

Most of an AI answer is built from other people's pages. In our study, a brand's own site supplied 2.9% of citations and other companies' sites 75.2%. Reddit and community threads alone drew 3.3%, more than brands' own sites. So give off-site work its own line in the plan, with an owner and a weekly target, not whatever time is left after publishing.

Your own pages carry the facts

Your site still matters, in a different role. Whoever writes the lists, threads and reviews that name you can only repeat what they can find about your pricing, features and who you serve. Keep those pages specific and current, because a stale fact on your site can end up on every page that copies it.

Step 5: Keep SEO Running, and Refresh Before You Add

Don't cut the SEO budget to pay for this

In Pew's March 2025 panel, 82% of Google searches showed no AI summary at all, and those still sent the clicks. The AI engines that search the web can only cite pages they can find and read, so crawlable, indexed pages serve both. We cover how the two fit in AI visibility and SEO.

Rework what you already have

Before commissioning new pages, look at the ones you have that cover your target questions. Put the direct answer in the first paragraph, add the facts only you have, and cut what you cannot source. Our page refresh checklist does this without inventing facts. A reworked page keeps the links and history it already has.

Step 6: Set the Budget Before the Tactics

There are three ways to resource this, and most teams mix them. Our GEO cost breakdown prices each one from sources you can open:

Decide what you lack first. A team that can write but cannot see what AI says needs measurement. A team that can see the gaps but has no time to close them needs hands. Set the monthly number, then pick the tactics that fit inside it.

Step 7: Measure Answers, Not a Score

Why a score misleads

An overall visibility score averages every question and engine you track. Between consecutive runs with no fix shipped, a brand's overall score moved up to 14.5 points in our tracking. Judge a quarter by that number and you will credit or blame the plan for noise.

Count the answers on the questions you worked

A single question's answers mostly hold still: in our study, 77.5% of brand, question and engine combinations either always or never named the brand across runs. So note which engines name you on each target question before you start, ask the same questions every week, and count. An answer that names you on two runs in a row is a win. Keep two questions you did not work on as a control: if they move as much as your targets, the engine changed, not your content. Our guide to telling whether a change worked has the full protocol.

A One-Quarter Plan You Can Copy

Month one sets the baseline, month two does the work, month three counts. Copy it into your planning doc and replace the questions with your own.

AI SEARCH CONTENT PLAN: ONE QUARTER

MONTH 1: SET UP
[ ] Write 20-30 questions buyers ask an assistant, each one whose answer could name vendors
[ ] Ask each one on ChatGPT and Gemini (and the other engines your buyers use), on two different days
[ ] Log: named or not, who is named instead, which pages are cited, the format of each cited page
[ ] Note the questions your rivals win and you do not: these are the plan

MONTH 2: FUND THE PAGES AI CITES
[ ] For each lost question, decide: your own page, someone else's page, or both
[ ] Own pages: publish or rework one page per lost question, in the format the cited pages use
[ ] Other pages: one outreach move a week (a best-of list, a thread, a review site)
[ ] Keep the SEO basics running on every page you publish

MONTH 3: COUNT
[ ] Re-ask the same questions every week on the same engines
[ ] Count answers that name you on each target question, against the month 1 baseline
[ ] Keep two questions you did not work on as a control
[ ] Keep what moved, drop what did not, and pick next quarter's questions

How Ranqo Fits the Plan

We built Ranqo to run this plan in one place, so each step above starts from data you already have.

• Steps 2 and 7. Ranqo asks your tracked questions every week on each engine your plan covers. Each question's insights page keeps every answer, so the baseline and the weekly count are already there.
• Step 3. When you generate ideas, Blog Ideas turns the questions your latest run left you out of into article ideas, with who AI names instead and the format of the pages it cites. Generate Blog drafts the post from the idea, at the length you pick, and can push it to WordPress as a draft.
• Step 4. Outreach lists the threads and pages AI cites in answers that leave you out, ranks them, and drafts a reply or a pitch when you ask.

The weekly runs keep asking the target questions after you publish, so the count in step 7 keeps arriving without anyone re-checking by hand.