One Person, 30 Days, Zero New Tools: A Playbook for Your First AI Citations

One person, 30 days, zero new tools. If you handed me a small business that never gets cited in ChatGPT or Perplexity, I would spend the first full week writing nothing.

Most small businesses do not have a content problem. They have a list problem. They publish pages and hope an engine picks them up without writing down the buyer questions that matter or checking who gets cited instead. I learned this running Google Ads: you do not rewrite ads before you pull the search term report. Here the report is 20 buyer prompts, run in both engines and logged citation by citation. Fifteen become the work list. Five stay untouched as a control, so the day-30 check has something to compare against. Find the gap first. Then close it.

1. Commit to the constraint before you touch the site

Cost: One person working in the gaps between real work, roughly an hour a day. Buys: A short list you can defend instead of a month of random publishing.

Cut the purchases the 40-item checklists tell you to make first: no visibility tracker, no content writer, no schema plugin audit. I used to tell clients to instrument everything before changing anything. I was wrong. For a small business with zero citations, measurement without a fixed prompt list produces prettier ignorance.

This is not a plan to become visible for every possible question. It is a plan to find the buyer questions where absence hurts, make a small number of changes, and check whether those questions move more than untouched ones. If a task does not serve that sequence, it does not get time this month.

2. Week 1: find the missing buyer questions

Cost: Three evenings, a spreadsheet, zero dollars. Buys: A list of prompts where buyers are close to spending and your business is absent. Cut: New writing. A page drafted before this check is a guess.

Start with 20 questions customers actually ask on sales calls, not the questions your blog already answers. Think price, turnaround, service area, how you compare with a named alternative, who the service is for, and who should skip it. Practitioner guides describe a similar exercise: write 10 to 20 real buyer questions, run them through AI search, and log who gets named versus who gets linked as a cited source. That distinction matters. A business owner asking, “Is ChatGPT recommending us?” needs to know whether the business appears in the answer, not just whether its site is somewhere in the footnotes.

Run all 20 in ChatGPT with browsing on and in Perplexity on the same day, logged out and in the same location. Screenshot each answer. Record every cited domain and mark whether your business appears at all. Treat ChatGPT and Perplexity as separate tests. In cross-platform studies, Reddit made up about 46.7% of Perplexity top-10 citations versus about 11% on ChatGPT, while Wikipedia accounted for about 47.9% of ChatGPT top citations in the source mix studied. Combine the engines into one score and you may work on the wrong gap.

Split the prompts before you change anything

Pick 15 prompts where you are missing and buyers are close to spending. Leave 5 similar prompts completely alone. Cost: the discipline not to fix everything you see. Buys: a comparison on day 30.

AI answers wobble. Across 1,765 B2B answers, all four models agreed on the same domain for the same question only 1.7% of the time. In another check, rerunning the same 72 prompts two days apart retained only 30% of URLs on Claude and 65% on Perplexity in a 13,184-citation analysis. Without untouched prompts, one flattering screenshot can pass for progress. I have seen enough ad-copy tests to know how that ends.

For each of the 15 work prompts, name the gap in one line:

• No relevant page exists.
• A page exists, but it does not answer the question directly.
• The answer exists, but an engine cannot readily read the page.
• The answer exists, but other sources do not mention the business.

Do not label all four problems “we need more content.” In an 8,000-citation study, ChatGPT cited vendor blogs under 3% of the time while Perplexity cited blogs in about 38% of cases with different source mixes. A missing ChatGPT citation may point to a third-party mention gap, not a shortage of adjectives on your blog. This four-gap breakdown goes deeper on separating content, crawl, and authority problems. Week 1 ends with 15 named gaps, not 15 post ideas.

3. Week 2: make the useful pages readable

Cost: One Saturday morning and a text editor. Buys: A better chance that an engine can find and read the answers you already have. Cut: New writing until you have checked the crawl basics. A polished answer hidden from a crawler is not a useful answer.

Check three things in order:

1. Crawler access. Open robots.txt and look for a stale Disallow: / left from staging. Check whether your rules block OAI-SearchBot, ChatGPT-User, or PerplexityBot; do not confuse those checks with decisions about training bots such as GPTBot and CCBot when reviewing crawler rules. The crawler readability check walks through the sequence if you want it beside you.
2. Bing indexing. Check whether your important pages appear there. If they do not, use the free Bing Webmaster Tools setup to verify the site and submit its sitemap; the practitioner setup guide also covers IndexNow as a way to flag new pages. ChatGPT Search can draw candidates from Bing, so ignoring that index while obsessing over Google leaves a gap. Submission is not a promise of a citation.
3. Visible page text. View Source on your five money pages. If the useful answer appears only after JavaScript runs, crawlers that do not execute it may miss the content even when the page looks fine to a visitor. Put key answers in the initial HTML rather than relying on a rendered view.

Keep one clear, canonical page for each question you decide to pursue. Do not create another page to compensate for one that is blocked or empty in source. That is how a small site turns a fixable technical problem into a larger editorial one. Week 2 is for making answers accessible, not making the page count bigger.

4. Week 3: answer the five best gaps first

Cost: Five pages, one per gap. Buys: Material an engine can quote and a buyer can use. Cut: The other ten work prompts for now, a full blog redesign, and any page that tries to answer three buyer questions at once.

Pick the five gaps closest to a buying decision. Where a useful page already exists, improve that page rather than publishing a second answer to the same question. Give each page a job: answer one prompt clearly enough that a reader does not have to hunt through an introduction to find it.

The writing order is simple:

1. Put a 40- to 60-word direct answer in the opening paragraph.
2. Use question-form H2s, with the answer at the start of each section.
3. Follow with the specific evidence a buyer needs: numbers, service area, turnaround, or a named comparison where it belongs.

There is a reason to be specific, though not a reason to pad. Cross-platform signal studies report more citations for content over about 2,900 words and visibility lifts associated with statistics, expert quotes, and author authority in the material studied. Those findings do not mean this small business needs five bloated essays. They mean a bare claim gives an engine and a buyer less to work with than an answer supported by details.

Say you install heat pumps in brick rowhouses in Baltimore for $14,500 to $18,000 in two days, and name the zip codes you serve. Do not say you provide efficient comfort solutions and expect anyone to work out the price, place, and timing. Use real details from the business, not numbers invented to make a page look quotable.

I used to tell clients to write for readers and let rankings follow. I was wrong in this narrow case. Write the first sentence so an engine can lift it cleanly. Write everything after it for the human who clicks. By the end of week 3, five buyer questions have direct, readable answers.

5. Week 4: get named where buyers already look

Cost: Outreach and corrections, not another writing sprint. Buys: A chance to appear in the third-party sources your own site cannot supply. Cut: Guest-post pitches and ten more blog posts.

Cross-platform signal data puts a large share of AI citations on off-site sources and reports a stronger correlation between brand web mentions and AI visibility than between backlinks and that visibility in the data studied. Correlation is not a placement guarantee. It is a good reason to check which outside pages already show up for your 15 prompts before you spend the week writing for your own site.

I call this getting named on pages the engines already read. The deck calls it authority building. Fine. Open your week-1 screenshots and make a short outreach list from the sources that recur. Look for:

• A directory profile with an old phone number or an incomplete description.
• A review-platform page, such as Clutch, that does not yet make your fit clear.
• A comparison post that names competitors but omits your business.
• A relevant forum thread where someone asked for recommendations.

Correct what you control. Ask for an accurate update where you do not. Do not pretend a comparison publisher owes you inclusion, and do not drop a sales pitch into an old thread just to manufacture a mention. Aim at the five most relevant opportunities; three useful corrections beat ten posts nobody asked for.

The source mix depends on the buyer question. In one B2B weekly run, Reddit accounted for 0% of ChatGPT citations and 3% of Perplexity citations, while niche listicles, review platforms, and comparison posts took more of the space in that B2B citation data. That is why I would check the actual answers before deciding that every business needs a Reddit plan. This engine-by-engine comparison is useful before you choose where to spend the outreach time. Week 4 is about relevant mentions, not mentions for their own sake.

6. Day 30: rerun the prompts, including the untouched five

Cost: Two hours with the same spreadsheet. Buys: A read on whether your work prompts moved differently from the control prompts. Cut: Victory screenshots without a baseline.

Rerun all 20 prompts in both engines under the same conditions as week 1: logged out, same location, browsing on for ChatGPT. Record the citations and named businesses again. Keep ChatGPT and Perplexity in separate columns. Then compare the 15 prompts you worked on with the 5 you left alone.

Look for treatment moving while control stays flat. Say the worked prompts go from 2 cited to 6 cited while the untouched prompts stay at 1. That is a reason to investigate what worked, not proof that every new citation came from your edits. If both groups move together, be wary of calling it your win. I have watched PPC tests go wrong the same way: someone celebrates a CPA drop that happened in the control too because demand shifted.

Expect the engines to disagree. Across 1,429 tracked answers to 88 contractor questions, engines cited almost entirely different domains. LinkedIn appeared 246 times in Perplexity versus 6 in ChatGPT; YouTube appeared 214 times in AI Overviews versus 4 in ChatGPT in that cross-engine source data. “Publish on LinkedIn for AI visibility” is not a universal conclusion from those numbers. Report each engine separately, and compare work prompts with control prompts before you call the month a success.

7. When the 30-day constraint lifts, stop doing the loop by hand

The one-person, zero-new-tools rule ends here. I would stop hand-checking 20 prompts every Friday and stop writing one-off pages from memory. That continuing loop is what I would hand to groas: ongoing search execution with a human responsible for direction, rather than a monthly scramble between real jobs. I would keep the buyer list myself. Ten minutes a week is enough to add the exact questions prospects asked on calls and delete prompts that never lead to revenue. Machines should watch the answers. You should own the questions.

After 30 days, you should know which buyer prompts you worked on, which untouched prompts moved anyway, and whether ChatGPT or Perplexity names you where money is on the line. Keep the list current, keep a readable answer for each question worth pursuing, and keep checking the outside pages buyers encounter. Then let continuous execution handle the weekly grinding the first month exposed.

This will not work for everyone. If buyers choose your business through referrals or foot traffic and never ask an engine before they call, spend that hour a day on callbacks and reviews instead. For everyone else, the list is the asset you keep when the 30 days end.