One person. Two hours a week. Thirty days. If a small business cannot make progress toward its first ChatGPT or Perplexity citation inside that constraint, an AEO retainer is not the first thing I would buy.

I would start with the few pages buyers ask about, the third-party sources engines already cite, and a monitoring loop the owner can run. No sitewide rewrite. No new software. No claim that a citation will appear on schedule: these answers change with the question and the session. The point of the month is to stop guessing, put quotable answers where they belong, and learn whether the work is small enough to keep in-house.

The constraint: two hours a week, $0 in new software

Each week gets one job and a hard stop at two hours. The four jobs are to measure, fix pages, work on third-party presence, and build the repeatable check. A sitewide audit, blogging calendar, 200-URL schema project, and new subscription do not fit. Cutting them is the plan, not a promise to catch up after dinner.

Paid monitoring starts at $29 a month for Otterly’s 15-prompt tier and climbs to $399 or more for Profound. Entry tiers can also limit the engines you see. You do not need one to learn which buyer questions produce a mention, a linked citation, or neither. A sheet, a logged-out browser, and the free versions of ChatGPT and Perplexity will do for month one.

Cost against the constraint: four two-hour blocks, no new software spend. What that buys is a small, measured set of decisions. It does not buy comprehensive coverage of every prompt or every engine.

1. Week 1: spend two hours finding the prompts worth fixing

Write 20 prompts your best buyer might actually type. Leave your brand name out. Use category and problem wording: “best drain plumber in [town] for weekend callouts,” “how much does [service] cost for a 3-bed house,” or “[product] vs [competitor] for [use case].” A brand-name prompt can tell you whether an engine knows you exist. It is less useful for finding the question where a buyer discovers somebody else.

Run each prompt once in ChatGPT and once in Perplexity, logged out and in a fresh session. In your sheet, record the engine, exact wording, brands named, sources linked, businesses recommended, and the domains supplying the answers. Keep mention, citation, and recommendation in separate columns. Being named is not the same as being linked as a source; neither means the answer told a buyer to hire you. That distinction follows a manual tracking method built around 10 to 20 stable prompts.

Give the runs 60 minutes and sorting the sheet 30. Use the last 30 to repeat two or three questions closest to a sale, including a small wording variation. Answers are non-deterministic. One operator was cited for a specific Supabase error query but not for a broader AI-app repair query the same night; a single word could change the result. That is a reason to keep the wording in your sheet, not a reason to run prompts until you get an answer you like.

Rank the five questions where a relevant answer cites someone else and you have a plausible page or third-party source that could answer it. “Flippable” does not mean easy or guaranteed. It means the question is specific enough to work on and close enough to the business to matter. Beside each one, note which domains the engines linked. If the same directory or discussion keeps appearing, carry that observation into Week 3.

What the two hours buy: a baseline you can re-run and five questions to focus the next three weeks. What gets cut: rank trackers, content briefs, and edits made before you know which answers you are trying to change. If you cannot name your five questions, do not start rewriting pages.

2. Week 2: spend two hours making three pages quotable

Match those five questions to no more than three existing pages. For many small businesses, that will be a main service page, a pricing or FAQ page, and a proof page with reviews or job photos. Pick the pages that can honestly answer the questions in your sheet. Do not force a page to cover a service or price it does not represent.

Work on the top of each page. Put the direct answer in the first two sentences, in plain language, followed by the details a buyer needs: who the offer is for, where it applies, what it costs when a price can be stated, and how fast it works when that is relevant. The passage-first method for AI Overviews and ChatGPT gives you a useful shape: one question, one 40- to 60-word answer, then the supporting detail below.

There is a mechanism here, not a magic word count. An answer near the top is easier to extract without making the engine infer your offer from a brand story. Perplexity’s reranker favors extractable answers; in one small test of 40 pages, 9 of the 11 pages ChatGPT referenced put a plain answer in the first 100 words. Neither observation promises that your page will be cited. Both are good reasons to stop burying the answer.

“World-class solutions” gives a reader or an engine nothing to repeat. A concrete price, service area, or scope does. That is the useful part of making copy quotable rather than clever. I used to tell clients to lead with the brand story. I was wrong. Answer the question first; let the story earn its place afterward.

Set a timer for 40 minutes per page. Read the opening against the target prompt, replace vague claims with facts you can stand behind, and leave the rest alone. If pricing depends on the job, explain that plainly rather than inventing a tidy number for the sake of a citation. Do not spend the block debating a redesign.

What the two hours buy: three pages that can supply a direct answer without guesswork. What gets cut: the rest of the site, new blog posts, and pages with no connection to your five questions. Three quotable pages beat thirty polished ones that dodge the question.

3. Week 3: spend two hours on sources other than your site

Your pages are only part of what an engine can use. In browsed ChatGPT answers, brands can be named more often than they are linked, and only a portion of retrieved pages become citations. Directories, discussions, and other third-party pages therefore matter even when your own copy is sound. The job is not to scatter your name across the internet. It is to look at the Week 1 sheet and work where relevant answers already come from.

Start with the five domains that appeared most often beside your target prompts. Which ones can legitimately include your business? A category directory may accept a listing. An existing profile may need its hours or service area corrected. A forum thread may warrant an answer if you have firsthand job knowledge. A cited comparison page you cannot honestly join is not a task for this week. Cross it off.

Use the block in that order. Spend the first 30 minutes checking the source list and choosing the accessible opportunities. Put the next hour into accurate listings or profile corrections, starting with Google Business Profile, Bing Places, or relevant category directories where they fit the answers you saw. Use the final 30 minutes for one additional action: submit the site to Bing Webmaster Tools if it needs doing, answer one relevant discussion without a pitch, or check whether OAI-SearchBot is blocked in robots.txt. It is separate from GPTBot, so do not assume one setting covers both.

Small storefront beside links to review sites and forums

Keep your name, address, and hours consistent wherever you make a change. Do not create a directory profile solely because a generic AEO checklist named the directory; your sheet is the reason to spend time there. The same rule applies to Reddit. If it appears in a relevant answer and you can contribute something useful, answer. If not, manufacturing a mention wastes the block.

What the two hours buy: a better chance of appearing in sources the engines already use for your questions. What gets cut: link-building campaigns, press outreach, and Wikipedia pages you do not qualify for. If a cited source has no honest path for your business, move to the next one. Do not argue with the engine’s source list; use it to set priorities.

4. Week 4: spend two hours making the check repeatable

Do not use the last block to declare victory after one favorable answer. Re-run the same questions, then make the sheet easy enough to use next month. The manual $0 approach becomes less work once the prompts and columns exist. Keep this month inside the software constraint: a Perplexity Sonar API check may cost well under $1 for 20 prompts, but it is still new spend, and setting up a script is another job. Leave it out.

Here is the Week 4 checklist. Run it in order and stop at two hours; the point is to preserve comparable answers, not to turn monitoring into a second marketing department.

  1. Re-run the same 20 prompts in ChatGPT and Perplexity (60 minutes). Use the same logged-out setup and fresh sessions; repeat the two or three money prompts if time allows.
  2. Record the date, engine, exact wording, brands named, sources linked, businesses recommended, and winning domains (20 minutes). Do not merge different questions into one row.
  3. Compare the new rows with Week 1 (20 minutes). Mark a No-to-Yes citation flip separately from a new mention or recommendation.
  4. Check the three edited page openings and the listings you changed (10 minutes). Confirm the answers and business details still say what you intended.
  5. Choose next month’s one page or source to revisit (10 minutes). Base it on a target prompt, not on whichever dashboard looks busy.
  6. Record the linked source, not just the brand name. This is the check owners skip: without it, a mention can look like a citation, and you can spend the next month fixing the wrong page.

The first five items allocate the two hours; the last is a rule for recording them, not another project. Refresh a page’s updated date when you make a real change, not to manufacture freshness. Leave a sitewide schema project out of this month. If your CMS strips markup or hides answers from crawlers, that is a different problem from writing three better openings.

What the two hours buy: a monthly diff you can act on. A new linked citation on a target question is the clearest signal for this playbook, but a mention or recommendation is still worth recording. What gets cut: daily checking, paid dashboards, and victory laps based on one flattering run. The next move comes from the pattern, not the best screenshot.

Where the two-hour plan runs out of road

This works while the question set, engines, and locations stay small. At 50 prompts, running each one twice across ChatGPT and Perplexity means 200 answers to read and score before you repeat any borderline result. Add more engines and locations, and monitoring can consume the entire block. Once that happens, nothing gets fixed. Check the clock in month two: if all you do is track, DIY has stopped paying for itself. The AI Overviews, ChatGPT, and Perplexity comparison is useful when you need to choose which engine to prioritize rather than covering all of them badly.

The other break is technical debt this plan deliberately leaves alone: JavaScript that hides prices from crawlers, duplicate location pages, a CMS that strips FAQ schema on publish, or review profiles split across two names. One operator’s fix order was answer-first pages, off-site sources already cited, schema, then re-measurement. That order assumes changes actually stay published. If they do not, or you need 40 pages fixed rather than three, two hours is no longer a useful constraint.

Skip this DIY version if you run multiple locations, carry hundreds of SKUs, or need to check four engines every week. That workload calls for continuous execution and a named human who owns the outcome. It is the distinction groas makes: autonomous work on search, with a strategist setting direction and guardrails, rather than asking an owner to live in a Friday sheet.

Start this Friday. Block two hours, run the 20 prompts, and fix nothing until the sheet tells you which five questions deserve the work. If a citation appears, you will know which question and source produced it. If it does not, you will still have a measured next move instead of an agency-sized guess. When the constraint lifts, keep the sheet and hand it to whoever takes over—your team or groas. The prompts remain the scoreboard. Your job changes from doing every check to checking what changed.