This is the practical companion to how ChatGPT decides which local business to recommend. That post explains the machinery; this one is the to-do list, in the order I would actually do it, including the popular steps that are not worth your time yet.
Nothing here requires an agency. Most of it is a few hours of unglamorous work.
Week 1: the foundation nobody wants to do
1. Get into Bing's index, genuinely, the first thing
ChatGPT has no web index of its own. When it needs live information it discovers candidate pages through Bing's search API, then fetches selected ones. Which means the question "how do I show up in ChatGPT" has an unglamorous first answer: be findable in Bing.
Almost every local business has spent a decade optimizing for Google and has never once looked at Bing. The front door is crowded and you have been queuing at it. The side door is open.
Do this: create a free Bing Webmaster Tools account, verify your site, submit your sitemap, and check that your key pages are actually indexed. You can import your Google Search Console setup in a couple of clicks. Twenty minutes, no cost, and it moves you from "not a candidate" to "a candidate".
2. Make one entity out of your scattered listings
Answer engines have to resolve you to a single confident entity before they will describe you. Conflicting information is not neutral. It is a reason to name someone else.
Pull up every place your business appears: Google Business Profile, Bing Places, Apple Business Connect, Yelp, your industry directories, your insurance or association listings. Make the name, address, phone, hours, and service area byte-for-byte identical. Kill dead listings for old suites and tracking numbers.
3. Make sure a fetcher can actually read you
A page selected for live fetching still has to return something useful. Check
that your key pages render their content in the HTML, not only after
JavaScript executes; that they load quickly; and that your robots.txt does not
block the AI crawlers you want, GPTBot, OAI-SearchBot, ChatGPT-User,
PerplexityBot, Google-Extended.
That last one catches people out. Some sites blocked AI crawlers wholesale in 2023 out of caution, and are now invisible by their own configuration.
Weeks 2–3: make your pages extractable
4. Write one page per real question
Fan-out means you are judged against many narrow sub-questions, not the broad one. So stop thinking in services and start thinking in questions. For most local businesses the highest-value set is:
- Cost: one page per major service, with a real range for your city
- Insurance / eligibility / payment, every plan or option, by name, as text
- Emergency or urgency, your actual availability and what to do now
- Comparison, the "X vs Y" decision your customers agonize over
- How to choose, the criteria, which conveniently are your strengths
5. Lead with the answer
The single highest-leverage edit you can make to an existing page: move the answer to the first sentence.
Pew found the median AI summary runs about 67 words, assembled from three or more sources. Your realistic share is a sentence or two. If your answer arrives in paragraph five, it will not be the sentence chosen.
Bad: "Every roof is different, and many factors affect pricing. Our team has served Austin for 25 years…" Good: "A full roof replacement in Austin typically runs $9,000–$22,000, depending on square footage, pitch, and material."
6. Make every claim survive being lifted out
Write in units that stand alone. A sentence that begins "As mentioned above" or "It depends" cannot be quoted. Put the number, the plan name, the hours, the service radius directly in the sentence.
7. Add schema markup, with realistic expectations
Mark up your organization, local business details, FAQs, and articles. Structured data is an acknowledged input on Microsoft and Google surfaces when selecting citation candidates.
Be clear-eyed, though: for ChatGPT and Perplexity there is no equivalent public confirmation, and there are no peer-reviewed studies on schema's effect on LLM citation. It is cheap, low-risk, and probably helps. It is not the lever some vendors sell it as.
Week 4 onward: corroboration you do not control
8. Get reviews that say specific things
Engines quote reviewers, and weight reviews containing concrete service language, procedure names, and geography above generic praise. Stop asking for reviews and start asking for specific ones, after a particular service, ask the customer to mention that service.
9. Be present where the engines already look
Reddit alone accounts for roughly 21% of citations in Google's AI Overviews, with YouTube close behind. Local news, trade associations, and community publications play the same corroborating role.
You cannot fake this and should not try. You can answer questions honestly in a local subreddit, keep your professional association listing current, and be the business a reporter can actually reach.
What not to bother with yet
Being told what to skip is worth as much as being told what to do.
llms.txt. It is not an official standard, adoption among the engines that actually cite is thin, and no major engine has confirmed using it. Costs ten minutes if you want a lottery ticket; do not let it displace anything above.- Publishing AI-written volume. Fan-out rewards specific, verifiable answers. Generic generated pages fail the filtering step precisely because they contain nothing checkable.
- Keyword density. You are being read semantically and quoted selectively. Repetition does not help you get lifted.
- Chasing every new engine. Google AI Overviews, ChatGPT, and Perplexity cover the overwhelming majority of the behaviour. Get those right first.
The sequence, if you want it as a plan
Days 1–7, be a candidate. Bing Webmaster Tools, listing consistency, crawlability and robots check. Almost no writing.
Days 8–30, be quotable. Rewrite your top three pages to lead with the answer. Publish your cost page and your insurance or eligibility page. Add schema.
Days 31–90, be corroborated. Specific review requests as a standing habit, directory and association profiles current, one comparison page and one "how to choose" page. Then measure.
How you will know it worked
Not from traffic. Citations convert at roughly 1% click-through, so a working AEO programme can look flat in analytics while it is succeeding.
Measure it directly instead: take the ten questions your customers actually ask, run them across ChatGPT, Google, and Perplexity on a schedule, and track how often you are named. Because answers are generated fresh each time, one check is noise, the rate over repeated samples is the signal. Watch the direction, and watch who else keeps appearing.
FAQ
Do I need to pay OpenAI to appear in ChatGPT?
No. There is no paid placement in ChatGPT's organic answers. You become a candidate by being discoverable through the search infrastructure it queries and readable to its fetcher, and you get cited by being the clearest available source for the question.
How long does it take to show up?
Indexing and listing corrections propagate in days to weeks. New pages need to be crawled and then earn enough corroboration to be trusted, so allow a month or two before a new cost or insurance page reliably appears in answers. Review language shifts how you are described within weeks of volume arriving.
Is Bing really that important?
For ChatGPT, yes. It discovers candidate pages through Bing's search API rather than maintaining its own index, so a site that is unindexed there is frequently not in the running. It has no bearing on Google AI Overviews, which use Google's own systems. Verifying in Bing Webmaster Tools is free and takes about twenty minutes.
Should I create an llms.txt file?
Probably not yet. It is not an official standard, adoption is thin among the engines that actually cite sources, and none of the major engines has confirmed using it. If you have spare time it costs little, but it should not displace getting indexed in Bing or rewriting pages to lead with the answer.
Can I just block AI crawlers instead?
You can, and some publishers do, but for a local business it is
self-defeating. Blocking GPTBot, OAI-SearchBot, ChatGPT-User,
PerplexityBot, or Google-Extended removes you from the answers your customers
are reading. Check your robots.txt, because a cautious block added in 2023 may
still be there.
What is the single highest-impact change?
For most businesses, moving the answer to the first sentence of the pages that already exist. It costs an afternoon, requires no new content, and directly addresses the filtering step where most pages are discarded after being retrieved.
The short version
Get into Bing. Make your listings agree. Make sure a bot can read your page. Then
write one page per real customer question, lead each with a specific answer that
survives being quoted out of context, and build corroboration you do not control.
Skip llms.txt and generated volume. Measure citation rate, not traffic.
The live demo shows what that measurement looks like for a real business, free, no signup. If you want the theory behind this list, read how ChatGPT decides; for the strategic framing, see AEO vs SEO.
Sources: Pew Research Center, Clicks and AI summaries · Search Engine Land, How schema markup fits into AI search, without the hype · Search Engine Land, Query fan-out guide