How to show up in AI search (ChatGPT, Perplexity, AI Overviews)

What AI search actually means for a small business

AI search is not one product. It is four or five different behaviours that all end the same way: your prospect gets a written answer with a short list of sources instead of a page of links to sift through. Google shows an AI summary above the normal results for a large share of question-shaped searches. ChatGPT runs live web searches and answers with footnotes. Perplexity is built entirely around that pattern. Gemini and Copilot do their own versions. Each one assembles its answer differently, but every one of them is reading pages that were found by ordinary crawling first.

That last point is the whole foundation, so it is worth stating plainly. There is no separate AI index you can submit to. These systems retrieve from the same web that search engines have always crawled, then summarise what they retrieved. A page that is not crawlable, not indexed, or not relevant to the question cannot be cited, no matter how good the writing is. Everything in this guide sits on top of the ordinary work of having a site that search engines can read, which is why the overlap with local SEO for small businesses is close to total.

What genuinely changes is the shape of the win. In classic search, you rank at position four and hope your title is compelling enough to earn the click. In an AI answer, there is no position four. There is a paragraph that names two or three businesses, or describes two or three approaches, with links attached. You are either in the paragraph or you are not, and the gap between those two states is much wider than the gap between position four and position five. That is the real argument for paying attention to this now rather than in two years.

The second change is who arrives. A visitor who clicks through from an AI answer has already read a summary that described what you do and, often, recommended you for their specific situation. They arrive further down the decision path than somebody who clicked a link out of curiosity. Fewer visits, warmer visits. That is why measuring sessions alone will mislead you here, and why the website ROI question has to be answered on leads rather than traffic.

How answer engines choose which sites to quote

Nobody outside these companies knows the exact selection logic, and anyone who tells you otherwise is selling something. What is observable, consistently, across all of them is a rough pipeline. A question comes in. The system rewrites it into several search queries. It retrieves a set of candidate pages. It reads the passages in those pages that look most relevant. It writes an answer from the passages it trusts most, and it attaches links to the pages those passages came from.

Every stage of that pipeline is a filter, and your site can fall out at any of them. Falling out at the crawl stage is a technical problem. Falling out at the retrieval stage is usually a content problem, because your page does not clearly address the question anyone actually asked. Falling out at the passage stage is a structure problem, because your answer is buried in the sixth paragraph next to three other topics and cannot be extracted cleanly.

1. Pages on your site 40 a typical small service site with service and location pages
2. Pages actually indexed 34 thin, duplicate, and blocked pages drop out here
3. Pages that answer a real question 12 the rest are navigation, legal, and general marketing
4. Pages retrieved against competitors 5 where your page is one of the best available sources
5. Pages quoted in an answer 2 the ones with a clean extractable statement

Illustrative shape for a 40-page service site rather than measured data. Run your own version from the Pages report in Search Console and a monthly spot check of your top questions.

The useful thing about seeing it as a funnel is that it tells you where to spend. If you only have six pages indexed, no amount of clever writing helps, because you have a crawling problem. If you have forty pages indexed and none of them answers a question a buyer would ask, you have a content strategy problem, which is the same problem covered in what pages a website needs. If you are being retrieved but never quoted, you have a structure problem, and that is the cheapest of the three to fix.

One more pattern is worth naming. Answer engines strongly prefer sources that corroborate each other. A claim that appears on your site, your Google Business Profile, two directories, and a local news mention gets stated confidently. A claim that appears only on your own site gets hedged or dropped. This is why the off-site work in section six matters more than most people expect, and why inconsistent business details do measurable damage.

The technical baseline: readable beats clever

Before any of the content advice is worth acting on, your pages have to arrive as readable text. A meaningful share of the crawlers feeding these systems do not run JavaScript. They request the URL, take whatever HTML comes back, and move on. If your headline, body copy, prices, and phone number only exist after scripts run in a browser, those crawlers see an empty shell.

Test this yourself in about a minute. Open one of your service pages, disable JavaScript in your browser settings, and reload. If the page still shows its heading, its copy, its price, and its contact details, you are fine. If you get a loading spinner or a blank frame, that is the first and most expensive problem you have, and it is a platform problem rather than a content one. Page builders and heavy template themes are the usual cause, which is one of several reasons the page builder comparison and the WordPress versus custom comparison both come down hard on server-rendered HTML.

The rest of the technical baseline is unglamorous and mostly familiar. Your robots.txt should not be blocking the crawlers you want. Your pages need unique titles and descriptions so the retrieval stage can tell them apart. Your internal links need to actually connect your pages so a crawler can reach the deep ones, because an orphaned page that is only reachable from your sitemap tends to get crawled rarely and treated as unimportant. Your site needs to load fast enough that crawling is cheap, which is the same argument made in Core Web Vitals explained and how to speed up your website.

Two smaller things matter more here than they do in classic SEO. First, put your facts in text rather than images. A price list rendered as a graphic is invisible to everything in this pipeline. So is a phone number baked into a header image, and so is an hours table that lives inside a screenshot. Second, date your content and keep it current. Answer engines lean toward sources that look maintained, and a page stamped with a date three years old competes badly against one updated last quarter. That is one of the quieter arguments for a real maintenance plan rather than a site nobody touches after launch.

• Main content readable with JavaScript off 60%
• Business name, address and phone identical everywhere 45%
• One page per specific buying question 35%
• Content dated and updated within the last year 40%
• Organization and FAQ structured data present 25%

Directional pattern across the small business sites I audit, not a published study. Your own site either passes each line or it does not, which is the number that matters.

Writing pages that are easy to quote

Once a machine can read your page, the question becomes whether it can lift one clean, correct, self-contained sentence out of it. That is a writing problem, and it is the part most small business sites get wrong, because the conventional marketing instinct is to build up to the point rather than lead with it.

Put the answer in the first two sentences. If the page is about what a bathroom remodel costs, the first two sentences should contain a number and a range. Everything after that can qualify, explain, and sell. A page that opens with three paragraphs about your family history and your commitment to quality has nothing extractable near the top, and extraction favours the top.

One question per page. A single page covering pricing, process, service area, and warranty answers all four questions badly from a retrieval point of view, because no passage in it is clearly about any one of them. Four pages, each built around one question, give you four independent chances to be the best available source. This is the same argument that drives service page separation in ordinary SEO, just with a sharper edge to it.

Write in complete, standalone sentences. A sentence that reads correctly when pulled out of context is a sentence that can be quoted. A sentence that starts with as mentioned above, or that depends on the bullet three lines earlier to make sense, cannot. This is a small habit that pays off constantly, and it also happens to make your copy clearer for humans. More of that thinking lives in website copy that converts.

State concrete facts, not adjectives. Fast turnaround is not quotable. Most repairs completed within two business days is quotable. Affordable pricing is not quotable. Standard drain clearing starts at 180 dollars is quotable. Adjectives describe how you feel about your business. Facts are what a machine can repeat on your behalf without risking being wrong, and machines are conservative about being wrong.

Use real question headings. Heading text that matches how a person actually phrases the question gives the retrieval stage an obvious handle. How much does a roof replacement cost in Orlando works better as a heading than Pricing, and it costs you nothing stylistically. A short frequently asked questions block at the bottom of every service page does the same job for the smaller questions that do not justify their own page.

Answer the comparison questions honestly. A large share of the questions people ask these systems are comparisons, and comparisons are the single most under-served content type on small business sites. A page that fairly explains when you are the right choice and when you are not gets cited far more than a page that claims you are always the right choice, because the second kind reads as marketing and the first reads as information. That is exactly why this site publishes a full set of custom versus template and do it yourself versus hiring a developer comparisons rather than pretending there is only one sensible answer.

Becoming a business a machine can describe confidently

There is a difference between a website with pages on it and a business a machine can describe. The second requires that a consistent set of facts about you exists in enough places that a system can state them without hedging. Name, address, phone, service area, service list, hours, and the person behind it. If those disagree across your own site, your Google Business Profile, and the directories that list you, every system reading them has to guess, and guessing usually means leaving you out.

The single highest-value fix here is boring: pick one canonical version of your business name, address, and phone number, then make every listing match it exactly. Not roughly. Exactly, including whether you use Suite or Ste, and whether the company name includes LLC. Then keep your Google Business Profile complete and current, because it feeds the local knowledge that these systems draw on for anything with a geographic component.

Structured data is the second layer. It is a block of machine-readable labelling that sits alongside your visible content and says which text is the business name, which is the address, which are the opening hours, and which are your question and answer pairs. It does not make anything cite you. What it does is remove ambiguity so nothing has to infer meaning from your page layout. For a small business site, three things cover most of the value: organization details on every page, local business details where you have a real service address, and question and answer pairs that match your visible text word for word. This page carries all three, and the FAQ answers below are byte-identical to the ones in its structured data, which is the part most implementations get wrong.

The third layer is authorship and accountability. Pages that name a real person, with a real role and a real way to reach them, are treated more seriously than anonymous content, and that gap appears to be widening. A visible about page with an actual human on it, a contact page with a real address and phone number, and bylines on anything substantial are cheap signals that separate you from the enormous volume of unattributed content these systems are learning to discount.

Your off-site footprint does most of the work

Here is the uncomfortable part. For most small businesses, what an answer engine says about you is shaped more by what other sites say than by what your own site says. Your site is one source. It is also the source with the most obvious incentive to flatter you, and these systems weight it accordingly.

Reviews are the clearest example. When somebody asks for the best option in their area, the answer is assembled largely from review platforms, directories, and any editorial coverage that exists, because those are the sources that look independent. A business with sixty recent reviews across two platforms gets named. A business with four reviews from 2023 does not, regardless of how good its website is. Asking every satisfied customer for a review is unglamorous, free, and probably the highest-return activity in this entire guide.

Directory and industry listings matter for a narrower reason: corroboration. Being listed consistently on the trade association site, the chamber directory, the licensing board register, and the two or three vertical directories in your industry gives a machine multiple independent confirmations of the same facts. You do not need eighty listings. You need the handful that are genuinely relevant to your trade, all saying the same thing.

Genuine mentions on sites that already have authority are the hardest to get and the most valuable. A local news piece, a supplier case study, a trade publication quote, a podcast appearance. Each one is a source that is not you, describing what you do. This is not link building in the spammy sense and no amount of paid placement on low-quality sites substitutes for it. One real mention beats fifty directory submissions.

1. Weeks 1–4

   Technical fixes land. Pages become readable without JavaScript, structured data goes in, business details get standardised everywhere they appear.

2. Months 1–3

   Indexing settles and questions containing your business name start returning accurate summaries. This is the first thing you will notice, and it is the easy half.

3. Months 3–6

   New question-shaped pages start being retrieved for non-branded questions. Reviews accumulated over the same period start showing up in recommendation answers.

4. Months 6–12

   Corroboration compounds. Being named consistently across reviews, directories, and third-party mentions is what moves you from occasionally mentioned to routinely recommended.

Measuring visibility when there is no ranking report

There is no Search Console for AI answers. No impressions chart, no average position, no export. That is genuinely inconvenient and it is also why so much nonsense gets sold in this space: without a scoreboard, nobody can be held to a result. You can still measure this. It just takes a spreadsheet and half an hour a month.

Build a prompt list. Write down the ten to fifteen questions a real prospect would ask before hiring you. Include your service plus your city, the price question, the how long does it take question, the is it worth it question, and two or three competitor comparisons. These are your tracked queries.

Run them monthly, by hand. Ask each question in ChatGPT, Perplexity, Gemini, and Google with AI answers on. Record three things per question: were you named, was a competitor named, and was the description of your business accurate. The third column is the one people skip and it is often the most actionable, because being described wrongly is a fixable information problem rather than a visibility problem.

Watch two automatic numbers alongside it. In your analytics, look for referral traffic from AI hostnames, which is small but real and growing. In Search Console, watch branded search impressions, meaning searches that contain your business name. When more people are being told your name in answers you cannot see, branded search tends to rise before anything else does. Setting both of these up properly is covered in website analytics 101.

Keep measuring leads, not sessions. If AI answers strip away your low-intent informational traffic while leaving your buying traffic intact, your session count drops and your lead count holds steady or rises. Judged on sessions, that looks like a disaster. Judged on leads, it is an improvement. Anyone measuring this on traffic alone will draw exactly the wrong conclusion, which is the same trap described in how to get leads from your website.

Half an hour a month, fifteen prompts, four engines, three columns. Three months of that is a trend line, and a trend line is the only form of evidence available here.

What is not worth paying for

This is a new enough category that the vendor market is running well ahead of the evidence. A few things are worth refusing outright.

Guaranteed placement in AI answers. There is no placement to buy. No submission endpoint, no partner program, no back door. Sponsored slots inside AI products are labelled advertising and are a different thing entirely from being cited as a source. A guarantee here is either a misunderstanding of how the systems work or a straightforward lie, and it is usually the second.

Monthly AI optimisation retainers with no deliverable. Most of the work described in this guide is one-time: fix the rendering, restructure the pages, standardise the business details, add the markup. On an existing site that lands in the $800–$1,500 range as a project, and a full rebuild where the site itself is the obstacle runs $2,800–$5,000 for a typical multi-page service site. What legitimately recurs is content, reviews, and maintenance. If a retainer cannot name what ships each month, it is not a retainer, it is a subscription. The pricing page and the cost estimator both show how that splits.

Bulk AI-generated content aimed at answer engines. Publishing two hundred thin pages to catch every question variation is the 2010 playbook with a new engine attached, and it fails for the same reason it failed then. These systems are specifically tuned to discount unattributed, low-substance, mass-produced content. Twelve pages that genuinely answer twelve questions outperform two hundred that gesture at two hundred.

Blocking AI crawlers on principle. For a business trying to get found, blocking the crawlers means opting out of the answers your prospects read. The content you are protecting is marketing copy you want people to see. There are real cases for blocking, mostly among publishers who sell access to writing, but a service business is not one of them.

Anything that requires you to stop building a normal good website. This is the summary test. If a tactic makes your site worse for humans, slower, less clear, or less honest, it is not going to earn you citations either. The systems are being tuned, continuously, toward the same thing a careful reader wants: pages that answer the question, state facts, and come from somebody accountable. Building that is what a custom-built site was always for, and it is why the answer to how to show up in AI search turns out to overlap almost completely with how to run a decent website.