One Person, Five Hours a Week, 60 Days: A $0-Software Playbook for AI Citations

One person. Five hours a week. Sixty days. Zero dollars for new software. Under that constraint, most advice about getting cited in ChatGPT and Perplexity is unaffordable: you cannot publish constantly, build authority everywhere, and still do your actual job.

I would do three things instead: find the twenty buyer prompts worth answering, fix the pages assistants are likely to fetch, and earn a few relevant mentions away from my own site. Everything else waits. Ahrefs found that only 12% of URLs cited by AI assistants rank in Google’s top ten for the same query, and 8% for ChatGPT. So I would not spend these sixty days trying to win a conventional top-three ranking before touching the pages that answer buyers’ questions.

The 40-hour budget: what gets cut before work begins

Five hours a week for eight weeks gives me 40 hours of work. I would spend ten on prompt research and a baseline, ten on access and priority pages, ten on two source pages, nine on off-site mentions, and one on a final check. The few days between week eight and day 60 are not a secret ninth workweek.

That budget rules out three tempting projects:

• Paid prompt tracking. Standard commercial plans run $97 to $189 a month. With twenty prompts, I would use a spreadsheet and spend the money elsewhere.
• An llms.txt project. SE Ranking found no statistical correlation between having the file and earning AI citations in its analysis of 300,000 domains. Under this constraint, it does not get an hour.
• A content calendar full of generic posts. Thirty broad articles will not answer a specific buyer’s pricing, comparison, or capability question as directly as two useful pages can.

The plan is narrow on purpose. I am not building an enterprise content operation in forty hours. I am making it easier for an assistant to find, understand, and cite the best answers this business can substantiate.

1. Weeks 1–2: find twenty buyer prompts and record the baseline

Cost: 10 hours total. Buys: a prompt list tied to sales conversations and a record of who gets cited now.

I would spend the first five hours in closed-won CRM notes, sales emails, and Google Ads search term reports. Not in a blank document trying to imagine what buyers ask. I want twenty commercial prompts that resemble questions someone might paste into ChatGPT or Perplexity shortly before a demo request or call.

I would look for three kinds of question:

1. Direct comparisons: “Brand A vs Brand B for ecommerce.”
2. Operational pricing: “Typical hourly rates for managed IT in Chicago.”
3. Capability checks: “Can Tool X handle multi-entity billing?”

“What is cloud computing?” can wait. A small business with five hours a week needs the narrower question where an answer could affect a purchase.

The next five hours go into a baseline. I would run all twenty prompts through ChatGPT with web browsing active and through Perplexity, then log five fields in a spreadsheet: Prompt, Brand Mentioned, URL Cited, Competitors Cited, and Winning Source Format. Mentioned and cited are separate columns for a reason. An assistant can name a business while sending the reader somewhere else for evidence.

I would note whether the cited page is vendor documentation, a directory, an editorial comparison, or a forum discussion. I would also record what the page puts near the top: a direct answer, a price, a table, a limitation. The point is not to declare that one engine always prefers one format. It is to see what happens on these twenty prompts before deciding what to build.

Keep at the end of week two: the twenty prompts, their baseline results, and the two questions where a better source page has the clearest job to do.

2. Weeks 3–4: remove access problems and rewrite the pages already doing the work

Cost: 10 hours total: four on access, six on page edits. Buys: reachable pages with answers an assistant can extract without digging.

I would spend four hours checking robots.txt, server responses, and any CDN, firewall, or security-plugin rules that affect search bots. OpenAI’s crawler documentation distinguishes GPTBot, which gathers training data, from OAI-SearchBot and ChatGPT-User, which serve search and live retrieval roles. Perplexity’s crawler documentation similarly distinguishes PerplexityBot from Perplexity-User.

Those distinctions matter if the business wants to restrict training access without accidentally blocking search access. I would check what the relevant bots can fetch, rather than assume that a permissive-looking robots.txt settles the question. A firewall challenge or 403 response can still get between an assistant and the page it needs. Fix the gate before polishing the room behind it.

The remaining six hours go to the homepage and primary service page. A look at live AI crawl logs puts those pages ahead of an ambitious new blog section for this pass. If either page opens with “transforming client potential into digital velocity,” I would replace that line with what the business does, for whom, and how.

I would use three rules:

• Answer first. Beneath each major heading, state the useful fact in the first two sentences. B2B practitioners discussing AI visibility make the same practical complaint I hear from buyers: nobody wants three paragraphs of scene-setting before the answer.
• Put available numbers in ordinary page text. Pricing baselines, turnaround times, minimum commitments, and compatibility belong where a visitor and a fetcher can read them, not only in a PDF or an interface that depends on JavaScript. If the business cannot substantiate a number, I would not invent one to fill a box.
• Use a plain comparison table when there is something real to compare. A table can make service-model differences easier to inspect than a row of decorative icons. It also forces me to state the trade-offs instead of hiding them under adjectives.

For the deeper server-response and page-structure work, I would use this AI-search readability checklist. I would not let that checklist expand this into a site rebuild. The deliverable is two clearer, reachable pages.

3. Weeks 5–6: publish two source pages, not a content calendar

Cost: 10 hours, five per page. Buys: direct answers to the two highest-value prompts from the audit.

I would choose the two contested commercial questions where the existing cited pages leave a gap the business can honestly fill. Then I would write one source page for each. No five-post sprint, no 2,000-word introduction to a topic the buyer already understands.

Each page gets the same working structure:

1. An H1 that matches the buyer’s question. Do not make the assistant translate a clever headline before it can use the page.
2. A direct answer in the first forty words. Give the scope and the relevant qualification there, not after a history of the industry.
3. A simple HTML table where comparison helps. Include only pricing, capabilities, timelines, or requirements the business can verify.
4. A frank account of the trade-offs. State who the offer fits, what it does not cover, and what changes the answer.

Say the prompt concerns implementation costs for boutique accounting software. I would not open with an essay on digital transformation. I would put the applicable cost information up front, explain what the scope includes, and list the technical requirements beneath it. If the business does not have a defensible range, the page should explain how its pricing is determined instead of borrowing a plausible-looking figure. A confident invented number is still invented.

This is where I would resist the urge to write for an imaginary algorithm. The buyer needs to understand the answer too. A clean paragraph, an honest table, and the limitation that matters to the purchase do more work than a page packed with variations of the same keyword. The page earns its place by answering one hard question well.

4. Weeks 7–8: build three relevant off-site footprints

Cost: nine hours total. Buys: up to three places beyond the company site where its details can be found and checked.

A company page is not the only kind of source an assistant may cite. In an evaluation of 6,847 Perplexity answers reported by the Machine Relations Index via AuthorityTech, editorial publications, vendor-owned domains, and community platforms all appeared among classified citations. I would use that as a reason to work beyond the website, not as a promise that any particular listing will produce a citation.

I would put these three items on the list, in this order:

1. Complete one relevant directory profile. I would choose a software or trade directory that appeared in the prompt audit, if there is one, and fill the fields the business can verify: pricing model, service region, integrations, and physical address where applicable. A half-empty profile gives a reader little to check.
2. Contribute to a discussion already visible in the audit. If a relevant Reddit or specialist-forum thread appears in the citation logs and allows a useful contribution, I would write from an operator’s perspective: costs, requirements, failure points, and where each approach fits. No drive-by pitch and no affiliate link disguised as advice. If there is no suitable thread, I would not manufacture one just to complete the list.
3. Work toward one ecosystem partner page. With a willing, non-competing tool or supplier, I would document a real shared workflow: what passes between the two services, what setup involves, and any applicable timeline. This depends on the partner. Nine hours buys the work, not somebody else’s approval.

Three useful footprints are the aim, not three links at any cost. I would rather leave the partner item unfinished than publish a vague mutual endorsement that tells a prospective buyer nothing.

5. Day 60: rerun the prompts and decide what actually changed

Cost: one hour from the 40-hour budget. Buys: a comparison with the week-two baseline, not a victory slide.

I would rerun the original twenty prompts through ChatGPT and Perplexity using the same wording and update the spreadsheet. Then I would separate three outcomes: the brand was mentioned, its own URL was cited, or a third-party page supplied the evidence. Those are different kinds of progress.

Four to six citations across the high-intent prompts is a target, not a guaranteed return on forty hours. If the count is lower, the sheet still tells me where to look next. If a directory is cited for a question and the business is missing there, the off-site work has an obvious next stop. If a competitor’s primary page is cited, I would inspect the answer it gives: is its scope clearer, its table more useful, or its pricing explanation easier to find? I would not call the engine deterministic and pretend one DOM change explains every result.

I would also keep the prompts that did not move. A page edit that improves the answer for a buyer but earns no citation is not evidence that the answer should be removed. It is evidence that I have not yet won that source slot. Measure citations without confusing them with revenue.

If five hours becomes two, protect the pages you have

A crisis can cut the weekly budget before day 60 or after it. At two hours, I would stop drafting new pages and pursuing new mentions. I would keep the smaller maintenance loop:

1. Ten minutes: Check logs or security rules for access problems affecting the relevant bots. Look for successful fetches and unexpected blocks.
2. Thirty minutes: Rerun one contested commercial prompt in both engines. If the cited source changes, inspect the replacement before changing your own page.
3. Eighty minutes: Update the pricing baselines, turnaround times, and technical specifications on the primary service page where the business has new, verified information.

That is 120 minutes. Trying to squeeze the entire forty-hour playbook into it would leave every task half-done.

When the constraint lifts, stop running the whole operation by spreadsheet

I would not spend a bigger budget publishing thirty generic posts. I would expand the prompt set, keep the commercial pages current, and connect citation work to the paid and organic search decisions the business already makes. At twenty prompts, I can watch the results by hand. At three hundred, while competitors change pages and paid-search campaigns keep moving, the spreadsheet becomes its own job.

That is the point where groas is the better fit than a traditional agency retainer or a monitoring tool that hands an already-busy person another checklist. It runs paid and organic search work continuously with specialized models, while a named human strategist sets direction and guardrails. Its flat monthly fee, no setup fees, and month-to-month terms also avoid paying for mechanical hours merely because an agency has put them on a timesheet.

Until there is room for that wider operation, I would hold the line: twenty buyer questions, reachable pages, two sharp answers, and a few off-site facts worth finding. Let the competitor spend its retainer on a topical authority map. I would rather give the buyer an answer.