A small firm improves its chance of being cited in AI answers by making its pages crawlable, stating the same facts everywhere, writing pages that answer client questions in plain sentences, earning real third-party mentions and testing with repeated prompts. Nothing guarantees a citation; these checks remove the common reasons a firm is skipped.
This is a checklist, not a tool review. Each item says what to check, how to check it in minutes and what a fix looks like. Google says that for its AI features ordinary SEO still applies, so a lot of this will already be familiar; the additions are the consistency audit and the prompt test, which most firms never do.
How the checklist is organised, and what it assumes
Citation depends on a chain: a crawler can fetch your page, the page is understood, the content is judged useful and trustworthy, other sources agree with it, and the tool decides to cite it for this question. A break anywhere ends the chain. The checklist has five groups matching those links, with 18 checks in all. Score each check 0 (fails), 1 (partly) or 2 (passes). A maximum of 36 is not the goal; the pattern of where you score 0 is.
Two assumptions. You run a professional or service firm with a website you can edit, and you want to be cited for a few specific services, not everything. And you accept that answers vary by wording, person and day: OpenAI notes that search results and citations can be incomplete, outdated or incorrect, which is why testing is part of the list. For how ChatGPT retrieves businesses, how ChatGPT finds a business and what a firm can control sets out the mechanism; the focus here is the checks themselves.
Group one: can the crawlers reach you? (checks 1 to 4)
- Robots file allows the search crawlers. Open yourfirm.example/robots.txt. Make sure no blanket "Disallow: /" applies to everything and that OpenAI's OAI-SearchBot is not blocked. OpenAI says each of its robots is an independent setting, so blocking GPTBot for training does not remove you from search.
- Firewall and bot protection. Ask your web host or agency whether a security setting blocks unknown bots. OpenAI recommends allowing its published IP ranges for the search crawler.
- Pages are indexable. Google says a page must be indexed and eligible to show in Search with a snippet to appear in its generative features. Search for
site:yourfirm.example/service-pageand confirm the page appears. Look for stray "noindex" settings left from a redesign. Then open Google Search Console, go to Settings and check the Search generative AI control: including the site is the default, and Google says an excluded site's links and content will not appear in AI Overviews or AI Mode. - Content is in the page text. If key facts live only in images, PDFs or an embedded widget, put them in plain page text as well.
Typical failure: a firm redesigned its site on a staging address, copied it live, and the "discourage search engines" setting came along. It scored 0 on checks 1 and 3, and no amount of writing would have helped until it was fixed.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowGroup two: do your facts agree everywhere? (checks 5 to 8)
AI tools often merge what several pages say about a firm. If they conflict, a tool may hedge, pick the wrong version or skip you. This group is the audit most firms have never done, and it costs an afternoon.
Build a facts sheet with the exact firm name, trading name, address, phone, opening hours, services, named professionals with their roles and credentials, and a 50-word description. Then compare it with each place that describes you.
| Source | Name | Address | Phone | Services | People |
|---|---|---|---|---|---|
| Your website footer | Match | Match | Match | Match | Match |
| Google Business Profile | Match | Match | Old number | Missing one | n/a |
| Professional body directory | Match | Old address | Match | Match | Former partner listed |
| Social profiles | Short name | Match | Match | Out of date | n/a |
| Review sites | Match | Match | Match | Match | n/a |
(Illustrative audit of a five-person architecture practice.) Five mismatches, found in about an hour, scored 1 on check 5 (name), 0 on check 6 (address), 0 on check 7 (phone), 1 on check 8 (services and people). Fix each at the source. For the profile, how to optimise your Google Business Profile with AI covers the listing side. Google also names Business Profiles and Merchant Center as ways to help local businesses and products appear in AI responses, which is a reason to keep the profile accurate.
Group three: are your pages answer-ready? (checks 9 to 13)
"Answer-ready" does not mean writing in a special way. Google says you do not need to rewrite content just for AI systems or chop it into small chunks. It means a stranger, or a model, can find a direct answer on the page without wading through marketing language.
- Each service page opens with a plain description. Two sentences: who it is for, what you do, how it is priced or scoped.
- Common questions are answered in full sentences. Fees, timescales, what you need from the client, what happens next.
- Specifics exist. A range, a process, a threshold, a typical duration, something a competitor's page does not say.
- The page shows who wrote it and who stands behind it. Named professional, role, qualification you can state accurately, last reviewed date.
- The page is the best one on its topic. Not duplicated across locations, and not thin. Google's guidance says non-commodity content with a unique, experienced point of view is likely to matter most in the long run.
A before and after of one page opening
BEFORE
"We are passionate about creating spaces that inspire. Our award-winning
team delivers bespoke architectural solutions tailored to your vision."
AFTER
"We are a five-person architecture practice designing house extensions
and loft conversions for homeowners. A typical project runs six to nine
months from first sketch to planning submission, and our design fee is
agreed as a fixed amount after a one-hour site visit."The second version is quotable. It says who, what, for whom, how long and how fees work. A model asked "how long does a house extension design take?" can lift the middle sentence, and a human can too. The first says nothing a different firm could not copy. (The practice, durations and approach are illustrative; use your own true figures.)
A quick way to test a page is to read the first two sentences to someone outside your profession and ask what the firm does. If they hesitate, rewrite. Do not use the tool to invent claims; use it to tighten your own true statements, and check the output for anything you did not say.
Group four: do others say the same about you? (checks 14 to 16)
Google says its AI features can show what is being said about products and services across the web, including blogs, videos and forum discussions, and ChatGPT search can cite many kinds of pages. So corroboration helps, but it must be real. Google's guidance specifically says seeking inauthentic mentions is not as helpful as it might seem.
- Reputable directories and professional bodies list you correctly. These are the easiest outside mentions, and you control the entry.
- You have genuine reviews on at least one major site. Ask real clients, never buy or write them. Answer reviews politely and factually.
- Someone independent has written about you. A local newspaper quote, a conference page, a partner's case study, an association newsletter. One real mention beats ten manufactured ones.
A practical route: pick two or three client-facing facts you can offer to a trade publication or association, such as a short guide to a recurring client question, written by one of your professionals. PR for AI search: how agencies earn citations in ChatGPT answers goes deeper on this approach.
Group five: do you test it, with a log? (checks 17 and 18)
- You run a fixed prompt set on a schedule. The same prompts, monthly, logged.
- You act on what the log shows. A wrong fact leads to a source fix; a missing service leads to a page change.
Build the prompt set from the questions clients actually ask, in four types:
| Type | Example for the architecture practice | What it shows |
|---|---|---|
| Name | "What does [firm name] do?" | Whether the model knows you and has facts right |
| Category | "Who designs house extensions for homeowners in [your area]?" | Whether you are named for the work you want |
| Question | "How long does it take to design a loft conversion?" | Whether your pages are cited as a source |
| Comparison | "How do I choose between an architect and a design-and-build firm?" | Whether your guidance appears in advice |
Run each prompt in ChatGPT and in Google's AI features on the same day, and record the result with a simple rubric. For each run, score 0 if the firm is absent, 1 if named without a link or source, 2 if named and your page is cited, and subtract 1 if any fact stated is wrong. Over a month of runs the average per prompt type gives you a stable view that one screenshot cannot.
Month 1 log (illustrative, 3 runs per prompt)
Name prompt: 2, 2, 1 -> avg 1.7 (address wrong once: -1 applied)
Category prompt: 0, 0, 1 -> avg 0.3
Question prompt: 1, 0, 2 -> avg 1.0
Comparison prompt: 0, 0, 0 -> avg 0.0Read the pattern. The firm is known by name but rarely appears when a stranger describes the need, so the work is in group three (answer-ready pages) and group four (outside mentions), not in group one. The comparison prompt scoring zero says nobody has seen the firm's advice on that decision, which suggests writing one page that actually compares architects with design-and-build firms, with a real opinion. For what Google itself can show you, Search Console and AI Overviews: what agencies can report explains the report, and Google's own guidance says to use its Generative AI performance report to see how content is performing in those features.
The main example scored end to end
To see the whole checklist working, take the illustrative five-person architecture practice and score it before and after a month of work. The scores are invented for the example, but the way they move is realistic.
| Group | Checks | Month 0 score (of max) | Month 1 score | What changed |
|---|---|---|---|---|
| Crawl access | 1 to 4 | 5 of 8 | 8 of 8 | Removed a site-wide noindex; allowed the search crawler through the firewall |
| Consistent facts | 5 to 8 | 2 of 8 | 7 of 8 | Corrected profile phone and services, directory address, former partner |
| Answer-ready pages | 9 to 13 | 3 of 10 | 6 of 10 | Rewrote openings on three pages; added fees and timescales |
| Outside mentions | 14 to 16 | 2 of 6 | 2 of 6 | Unchanged; takes longer, planned for month two |
| Testing | 17 and 18 | 0 of 4 | 4 of 4 | Prompt set built and run; action list written |
| Total | 18 | 12 of 36 | 27 of 36 |
The total rose from 12 to 27, but the interesting row is the one that did not move. Outside mentions are the slowest group and the hardest to fake, so the practice schedules an outreach task for month two rather than pretending the score is complete. Notice too that the checklist score and the prompt-test score are different measures. The checklist tells you what you have done; the log tells you what the tools do. Report both and do not let a good checklist total convince you that citations have followed.
A service page, filled in
An empty template is easy to nod at, so here is a filled one for a single page, again illustrative. It follows the check numbers above.
H1: Loft conversion design for homeowners
Opening (check 9): We design loft conversions for homeowners in houses with
a roof space of at least 2.2 metres at the highest point. Our fixed design
fee is agreed after a one-hour site visit.
Process (check 11): 1. Site visit and measured survey. 2. Two layout options
within two weeks. 3. Planning drawings. 4. Building-regulations drawings.
Typical timeline (check 11): six to nine months from visit to submission.
Questions (check 10): Do I need planning permission? What does the design
fee cover? Can I stay in the house during building?
Author (check 12): Written by [named architect], reviewed [month and year].
Facts line (checks 5 to 7): [Firm name], [address], [phone].Notice that the page carries a stated condition (the 2.2 metre roof height). Conditions are what turn a page from brochure copy into something a model can use to answer a conditional question. The questions block should have answers in full sentences, each one standing on its own, so an excerpt still makes sense when lifted out.
Using AI to audit your own page, without inventing anything
An assistant can help you audit pages against the facts sheet, provided you give it the true facts and ask it only to compare. Paste the page text and the sheet into a new chat, with a prompt like this:
Below are (1) my firm's facts sheet and (2) the text of one web page.
List every statement on the page that conflicts with the facts sheet, every
fact on the sheet that the page omits, and any sentence a stranger could not
understand without industry jargon. Do not rewrite anything and do not add
facts. Quote the exact words you are flagging.The output is a list of lines for you to fix, not copy. Typical findings are an old fee quoted in an old news post, a partner described with last year's title and a service named differently on two pages. One caution: do not paste private client material into a tool as part of this exercise. Your own website text and facts sheet are public, so they are safe to use; client documents are not.
A realistic mistake from a first attempt
A small bookkeeping firm asked an AI tool to rewrite its service pages to be "AI-friendly". The tool added phrases such as "trusted by hundreds of clients" and "industry-leading accuracy", neither of which the firm could support. Worse, it dropped the line that explained the fixed fee. The firm caught it because the owner compared the new text with the facts sheet, and restored the fee line. The lesson is that tools smooth and embellish; you decide which facts stay. Every sentence that remains must be one you could defend to a client or a regulator.
What to skip, because it eats time without evidence
- llms.txt as a fix. Google says Search ignores it. It may suit other tools you use, but it is not a citation lever you can count on.
- Special markup for AI. Google says there is no special schema needed for its generative features. Ordinary structured data is fine for rich results, as long as it states the same facts as the visible page.
- Dozens of near-duplicate question pages. Google's guidance warns that creating separate content for every variation primarily to manipulate responses can breach its scaled content abuse policy.
- Third-party tools that claim inside knowledge. Google says no third-party tool has access to its internal ranking or AI systems. A tool may still help you log prompts, but judge its claims accordingly; what a small firm should ask an AI visibility provider lists the questions.
A thirty-day plan with time estimates
| Week | Task | Time |
|---|---|---|
| 1 | Score all 18 checks; fix any group one failures with your web host | 3 hours |
| 2 | Build the facts sheet; correct the five most visible listings | 4 hours |
| 3 | Rewrite the openings of the three main service pages; add one questions page | 5 hours |
| 4 | Run the prompt set three times; score; decide next month's page or outreach task | 2 hours |
Fourteen hours in a month is a realistic load for a small firm, and most of it improves the site for human visitors regardless of what AI tools do. If the log barely moves after three months, the cost has still bought a cleaner site and consistent listings. Before paying for tooling, read what generative engine optimisation is and whether your business needs it, which puts the whole subject in proportion.
When a score refuses to move
If checks in groups one to three are passing and the category prompts still never name you, look at the competition on those prompts. Open the sources cited in the answers and see who they are: directories, publications, a handful of firms with long guides. If directories dominate, getting an accurate, detailed listing there is the quickest route. If competitors' guides dominate, you need a page that is more useful than theirs on a narrower question, written by someone who does the work. A firm in a crowded field may choose a niche (a client type, a situation or a service combination) where a page can realistically be the best answer. Narrowing is slower than hoping, but it is the one change that works whether the citing tool is ChatGPT, Google or something else.
Questions about improving AI SEO
Is AI SEO different from ordinary SEO?
Mostly it is the same work. Google says its generative features are rooted in its core ranking and quality systems, so standard SEO still applies. The extra effort goes into clear facts, consistent descriptions and pages that answer questions directly, plus testing what AI tools say about you.
Do I need an llms.txt file?
Google says its Search does not use llms.txt files and that creating one neither helps nor harms visibility there. OpenAI's pages I read do not describe the file as a ranking input. You can add one if a tool you use asks for it, but do not expect it to change citations.
Does structured data help me get cited?
Google says structured data is not required for its generative AI features and no special markup is needed, though it remains useful for rich results. For ChatGPT, whether its crawler uses markup is not stated on OpenAI's pages, so treat it as optional and put the same facts in plain text.
How soon will changes show in AI answers?
There is no fixed timetable. Crawlers revisit pages at their own pace, and OpenAI says search systems can take about 24 hours to adjust after a robots.txt change. Judge by a log of repeated tests over several weeks, not by one run the day after an edit.
Further reads
- How to Track Whether AI Search Is Sending You Customers — Check whether AI answers send you enquiries.
- Can Estate Agents Be Found in AI Search? What to Fix Now — A local-professional version of the same checks.
- How AI Overviews Are Changing Local Search for Small Businesses — How Google's AI answers change local search.
- AI SEO for Chiropractors: Local Pages and Showing Up in AI Answers — Local pages and reviews for another practice type.
- How to Monitor Brand Mentions in ChatGPT for a Client, Week by Week — A weekly schedule and log for ChatGPT mentions.
- Best AI Overviews Tracking Software: What a Local Business Needs — What AI Overviews tracking tools cost and show.
- SEO for AI Search: A Client Audit Checklist for Agencies — A deeper technical audit if you hand this to an agency.
- AI Schema Markup Generator: Validate Client Structured Data Before Launch — Generate and validate schema markup if you use it.
- Google AI SEO Agency: Questions to Ask Before You Hire One — A buyer's checklist for hiring a Google AI SEO agency: proof to demand, reports to expect, AI Overview tracking, contract clauses and red flags.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Search Central guide to optimising for generative AI features and Search Console help on the Search generative AI control; OpenAI crawler documentation and Help Centre (Searching the web with ChatGPT), read 4 November 2026. Example scores and firms are illustrative.