A search engine used to hand the buyer a list. An assistant hands back a summary, and your business either appears inside that summary or does not exist for that conversation.
That is the whole change. Everything else in this post is about what it means for a business that wants to be named when someone asks ChatGPT, Gemini, Perplexity or Google's AI Overviews a question that you should be the answer to.
We wrote a shorter version of this piece a year ago. Since then the tools have matured, Google has published its own guidance, and we have watched enough businesses move in and out of AI answers to say more precisely what moves the needle. This is the longer version.
What actually changed
The buyer no longer scans ten links and chooses. They read one answer that already made the choice for them, and they rarely scroll past it.
In the old model, being on page one was enough to get a fraction of the clicks. A buyer would compare three or four sites, form an opinion, and call someone. Your website did the persuading.
In the new model, the assistant does the comparing before the buyer sees anything. It reads the public web, decides which businesses are relevant and credible, and writes a short answer naming two to five of them. Your website still matters, but now its first job is to persuade a language model, not a person. The person comes second, after the model has decided you are worth mentioning.
This is not a small shift in traffic. It is a shift in where the decision happens. A business that is not in the answer does not get a chance to be bad at closing; it never gets the lead at all.
How assistants choose which businesses to name
Assistants pick businesses the same way a careful stranger would: they look for clear statements of what you do, evidence from other people that you do it, and consistency between every place your name appears. There is no secret ranking factor, and Google has said so plainly.
Google's own documentation for AI Overviews and AI Mode states: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It also says: "There's also no special schema.org structured data that you need to add." (Google Search Central, AI features and your website)
That sounds like nothing has changed. It has. The same document describes a technique called query fan-out: the system issues multiple related searches across subtopics and sources while it writes the answer, and pulls in supporting pages as it goes. So a single question from a buyer becomes several searches behind the scenes, and your business needs to show up well in enough of them to be worth citing.
Each assistant does this a little differently, and the differences matter.
Google AI Overviews and AI Mode
Google's AI features draw from Google's own index. To be eligible as a supporting link, a page must be indexed and eligible to appear in Search with a snippet. That means your existing search visibility is the floor. A page that does not rank in ordinary results does not suddenly appear in the overview.
For local questions, Google also leans on the Business Profile and the map pack. The three local ranking factors Google publishes are relevance, distance and prominence, and prominence is "based on info like how many websites link to your business and how many reviews you have" (Google Business Profile Help). AI Overviews for "best plumber near me" type queries inherit those same signals.
ChatGPT
ChatGPT with search enabled runs live web searches, reads the top results, and writes an answer. In practice it favors pages that state things plainly, pages from well-known directories and review sites, and pages it has encountered often enough to associate your name with your service. If your name appears on your site, on Google, on Yelp or Trustpilot or the local equivalent, and on a trade association page, all describing the same business in the same words, the model treats that agreement as evidence.
Perplexity
Perplexity cites its sources inline and tends to prefer pages that answer the exact question in the first paragraph. It is the most literal of the four. If a buyer asks "which commercial cleaning companies in Manchester do medical facilities," Perplexity rewards the page that says, in its first two sentences, that you are a commercial cleaning company in Manchester that cleans medical facilities.
Gemini
Gemini is wired to Google's index and to Google Maps data, so for local questions it behaves much like AI Overviews. For broader questions it behaves more like ChatGPT: it reads several sources and summarizes. Consistency across sources matters here most, because Gemini will often check your website against your Business Profile and against third-party listings before naming you.
The practical conclusion is the same for all four. You do not optimize for an assistant. You make your business legible to any reader that is scanning the public web for a clear, corroborated answer.
Answer-first pages
An answer-first page gives the answer in the first two sentences and spends the rest of the page proving it. That is the single most effective structural change you can make, and it costs nothing but discipline.
Most service websites open with an adjective. "Trusted. Reliable. Family-owned since 1987." None of that is quotable. An assistant looking for a business that repairs heat pumps in Tauranga cannot use a sentence that does not contain the words heat pump or Tauranga.
Compare two openings for the same page.
The weak version: "Welcome to Harbour Climate. We pride ourselves on exceptional service and customer care. Contact us today to learn how we can help."
The answer-first version: "Harbour Climate repairs and installs heat pumps for homes and small commercial buildings in Tauranga and the Western Bay of Plenty. Repairs start at NZ$180 for a call-out and diagnosis, and most installs are done within a week of the quote."
The second version is dull. It is also exactly what a model needs: what you do, who for, where, and what it costs to start. That is the sentence that ends up in the answer.
The same rule applies to every section of the page, not just the top. Each H2 should open with the answer to the question the heading implies, then explain. A heading that says "How long does an install take" should be followed by "Most installs take one day on site" before any discussion of why.
A few more things an answer-first page needs:
- A plain-language service list, one service per line, with a sentence each. Not a wall of icons.
- A starting price or a price range. Businesses resist this, and assistants reward it. "From $X" is enough.
- Service area stated in words, not only in a map embed. A map image is invisible to a model.
- A short "who this is for" paragraph. Residential, commercial, both. Small business, enterprise. Local, nationwide.
- Real hours, a real address if you have one, and a real phone number that matches everywhere else.
If you want to know where you stand first, book a call and we will walk through your site, listings and search with you.
Entity clarity: making your business one thing
Your business needs to be one unambiguous entity on the web, with the same name, address, phone number and description everywhere it appears. When those details disagree, a model has to guess which version is true, and it often resolves the doubt by leaving you out.
Think of it from the model's side. It finds "Harbour Climate Ltd" on your website, "Harbour Climate Heat Pumps" on Google, "Harbour Climate Services" on a directory with an old phone number, and a Facebook page with no address at all. Are these one business or three? A human would work it out. A model with a few hundred milliseconds to write a sentence will not bother.
Three things fix this.
Organization schema
Add Organization structured data (or LocalBusiness, which is a subtype) to your home page, and make it match the visible text exactly. Google's guidance is that structured data should reflect what a person can see on the page, so do not put a different name or phone number in the markup than on the page itself.
At minimum the markup should carry your legal or trading name, your logo, your URL, your address, your phone number, your service area, and a sameAs array. Google documents the Organization type here: Organization structured data. The full vocabulary is at schema.org/Organization.
Consistent NAP
NAP is name, address, phone. Pick one spelling of your name, one format for your address, one phone number, and use them identically everywhere: your site footer, your Business Profile, Apple Business Connect, Bing Places, every directory, every social profile, your email signature, your invoices. Not "similar." Identical, down to whether you write "Suite 4" or "Ste 4."
This is tedious work and it is the reason our HyperCare+ subscription includes keeping citations identical everywhere as a standing task rather than a one-off project. Listings drift. Someone updates the phone number on Google and forgets the three directories that copied the old one.
sameAs
The sameAs property in your Organization schema is a list of URLs that are definitively you: your Google Business Profile, your LinkedIn page, your Facebook page, your listing on the relevant trade body, your Wikipedia or Wikidata entry if you have one. It tells any reader "these profiles are the same entity as this website." It is the cheapest piece of entity disambiguation available and most small business sites do not have it.
Reviews and third-party corroboration
An assistant will not name a business on the strength of that business's own website. It wants someone else to have said the same thing. Reviews, directory listings, press mentions and trade memberships are that someone else.
This is where the old and new models overlap most. Google's local ranking documentation already said prominence depends on inbound links and reviews, and that "more reviews and positive ratings can help your business's local ranking." An AI answer is a layer on top of that ranking, so the same inputs feed it.
What is different is that assistants read the text of reviews, not just the star count. A review that says "they rewired our 1920s house in Leith and it passed inspection first time" tells a model three things: the service, the location, and the outcome. Forty reviews that say "great service, would recommend" tell it almost nothing.
You cannot write your customers' reviews for them, but you can ask better. When you request a review, ask the customer to mention what you did and where. Most people will, if asked.
Two other forms of corroboration matter more than they used to:
- Mentions on pages you do not control. A local newspaper story, a supplier's case study, a chamber of commerce listing, a "best of" roundup from a blog in your city. Each one is an independent statement that you exist and do what you say.
- Structured directory presence. Google, Apple, Bing, and whichever vertical directory dominates your trade and country. In the UK that might be Checkatrade; in Australia, hipages; in Germany, MyHammer; for B2B software, G2 or Capterra. Assistants trust these because they are consistent and machine-readable.
Replying to reviews is part of this. A reply is a second piece of text under your name, in your voice, confirming the service and location. It also signals to a reader, human or otherwise, that the business is attended. HyperCare+ answers every review within 1 hour for that reason.
What changes for local, national and global brands
A local business needs to win a map, a national brand needs to win a category, and a global brand needs to win the same category separately in every market it serves. The work scales in a specific way, and most of the extra work at each level is about disambiguation.
Local
If you serve one town or city, your Google Business Profile is doing most of the work and your website supports it. The checklist is short: answer-first service pages, one for each major service; a location page that names the area in words; Organization or LocalBusiness schema with sameAs; identical NAP across every listing; a steady flow of reviews that mention service and place.
Outside the US, Google Business Profile still dominates in most markets, but the supporting directories differ. Know which two or three listings a buyer in your country actually uses and make sure you are on them.
Multi-location
Each location needs to be its own entity. That means a separate Business Profile per location, a separate location page on your site per location, each with its own LocalBusiness schema, address, hours, phone and reviews. Do not merge them into one "locations" page with a list. A model cannot cite a list item.
The hard part is keeping the shared parts consistent (brand name, description, logo) while keeping the location-specific parts distinct. This is where businesses most often drift: three branches with three slightly different names.
National
A national service brand has no single map to win, so it has to win the category question: "which companies do X across the country." Those answers draw heavily on third-party comparison content, press coverage, and the clarity of your own category statement.
Your home page needs to say what category you are in using the words a buyer would use, not your internal brand language. If you call yourself a "property services partner" and buyers search for "commercial cleaning company," the model will not bridge that gap for you.
National brands also need a reviews strategy that is not purely local. Trustpilot, Google reviews aggregated at brand level, industry awards and trade press all feed the national answer.
Global
A global brand has to do the national work in every market it serves, and then make sure each market's version is clearly separated so an assistant does not serve the US page to a buyer in Singapore.
That separation is done with hreflang annotations, country-specific pages with local prices and local contact details, and local entity signals (a local Business Profile, local directory listings, local reviews) in each market. We cover the mechanics in International SEO: hreflang and multi-language sites, and Google's reference is here: Localized versions of your pages.
The mistake global brands make is treating translation as localization. A page in German with a US phone number and dollar prices is not a German entity. The model will notice the mismatch, and so will the buyer.
How to measure AI visibility
You measure AI visibility by asking the assistants the questions your buyers ask, on a schedule, and recording whether you are named, what position you are named in, and what the assistant says about you. There is no equivalent of Search Console for this yet, so the measurement is partly manual and partly tooling.
Here is the process we run for HyperCare+ clients, which you can run yourself at a smaller scale.
- Write down 15 to 30 questions a buyer would actually ask. Not keywords. Full questions, in the phrasing a person would use. "Who does emergency boiler repair in Leeds on a Sunday." "What's a good CRM for a 12-person landscaping company." "Best king cake bakery in Baton Rouge that ships."
- Ask each question in ChatGPT, Gemini, Perplexity and Google (with AI Overview or AI Mode). Use a clean session or a private window each time so previous conversation does not bias the answer.
- Record four things per answer: were you named; in what order; what did it say about you; which sources did it cite.
- Repeat monthly. Weekly if you are actively changing things.
Three patterns emerge quickly.
First, you will find questions where you are named and questions where a competitor is named instead. Look at what the assistant cited for the competitor. That is usually the page or listing you are missing.
Second, you will find questions where the assistant describes you wrongly: old address, discontinued service, wrong price. Trace that back to the source it cited and fix the source.
Third, you will find that Google's AI Overview and Gemini tend to agree with each other and with the map pack, while ChatGPT and Perplexity lean more on third-party pages. If you are visible in one pair and not the other, that tells you which layer to work on.
Alongside this, keep watching the signals that feed the answers: Search Console impressions and clicks for the pages that should be cited, map pack rankings for your core terms, review volume and content, and the number of indexed pages that mention your brand. These are leading indicators. AI answers are the lagging one.
If you want someone else to do this, map pack and AI-answer tracking across ChatGPT, Gemini and Perplexity is part of every HyperCare+ tier, reported in plain language monthly.
What this is not
It is not a reason to rewrite your site for robots, stuff pages with question headings, or publish an "llms.txt" file and wait. Google has said no new machine-readable files are needed. The businesses that win AI answers are the ones that are already clear, consistent and corroborated. The assistant is a reader, and it rewards the same things a careful human reader does.
It is also not a reason to abandon ordinary search. Google's AI features pull from Google's index, so the page that is not indexed, not ranking and not linked internally is not eligible to be cited. Ordinary technical SEO (crawlability, indexing, speed, internal links, schema that matches the page) remains the foundation. Our Systems page describes how we build that foundation into every site rather than bolting it on.
What to do this quarter
Do five things, in this order, and measure before and after.
- Baseline. Write your 15 to 30 buyer questions and run them through the four assistants. Record the results in a spreadsheet. This is your starting point and it takes an afternoon.
- Fix the first 200 words of your home page and each core service page so they answer "what, for whom, where, from what price" in the first two sentences. Remove adjectives that carry no information.
- Unify your entity. One name, one address format, one phone number. Update the Business Profile first, then your site footer, then every directory you can find yourself on. Add Organization or LocalBusiness schema with a
sameAsarray to your home page and confirm it matches the visible text. - Change how you ask for reviews. Ask customers to say what you did and where. Reply to every review, and in the reply confirm the service and location in your own words.
- Re-run the baseline at the end of the quarter. Note which questions moved, and which sources the assistants are now citing.
If you would rather hand all of this to someone, Mission ALPHA rebuilds the website, the entity signals and the listings in one 45-day project, and HyperCare+ keeps them maintained afterwards.If you want to know where you stand first, book a call and we will walk through your site, listings and search with you.
The buyer is still out there asking the same questions. The only thing that changed is who answers them first.
If this describes your business, one call will show you which part is costing the most.
