One Owner, Four Weeks, No Agency: Pick One AI Answer Engine and Get Citable

One owner. Four hours a week. Four weeks, no agency and no paid tools. That is not enough time to chase citations in Google AI Overviews, ChatGPT and Perplexity at once. Spend week one finding which engine’s pipeline you already half-feed, then give the remaining three weeks to that engine. Do not touch the other two this month.

I compared the engines in Google AI Overviews vs ChatGPT vs Perplexity: Which Should You Fix First?. Choosing one is only half the problem when you are doing this yourself. This second look puts that choice inside a month one person can actually run: diagnose, fix, measure, stop.

The budget is 16 hours total: roughly four hours a week for 30 days. Use the free tools you already have, including Search Console, your host or CDN logs, and your Google Business Profile. That buys a diagnosis and focused work on one engine. It does not buy three parallel projects.

Cut the site rebuild, the new blog cadence, the schema project, the 20-page content sprint and the review-blitz. I used to tell clients to fix everything at once. I was wrong. With four hours a week, breadth is how you spend 16 hours and change nothing an answer can quote.

1. Week one: find the engine you already half-feed

Cost: four hours. Buys: one target instead of three guesses. Work through the checks in order. Do not start rewriting pages because a log entry looks promising.

1. Check Google rankings (90 minutes). Open Search Console, filter the last 28 days, and write down the 10 queries that sent you the most clicks in your money category. Leave out branded terms; look for the questions buyers ask before they pick you. Search each in Google in an incognito window and note where your page sits. If three or more pages sit in positions 1 to 10 for those questions, your pages already feed the normal crawl-and-index path Google AI Overviews starts from. That is an opening, not a citation guarantee: only 37.9% of cited URLs in a March 2026 study of 863K SERPs also sat in the top 10. If none of your pages ranks in the top 20 for those 10 queries, Google is the slow road this month.

2. Check live fetches (90 minutes). Ask your host where raw access logs live, or open your CDN log view. Filter the last 30 days for ChatGPT-User, OAI-SearchBot, PerplexityBot and Perplexity-User. The names are not interchangeable. OpenAI uses OAI-SearchBot for search discovery and ChatGPT-User to fetch a specific page for a live question; PerplexityBot builds the index, while Perplexity-User fetches live and generally ignores robots.txt. Count fetches by URL, not just total hits. Our own homepage took 834 ChatGPT-User live-answer fetches in 30 days. That shows what live-fetch behavior can look like, not what your site must reach. If your pricing page or how-to guide takes 40 fetches while your blog takes zero, you have a candidate page.

3. Check what a fetcher can read (60 minutes). Open the top two candidate pages from those checks, view source, and search for the price, steps or verdict. If a key fact appears only after your site’s JavaScript runs, ChatGPT and Perplexity likely cannot read it in the fetched HTML. A Vercel and MERJ analysis of about 1.3B fetches found none of the major AI crawlers render JavaScript. Grade the pages you have; do not build new ones yet. Before you close the check, write down 10 to 15 buyer prompts and run a baseline in your likely engine. Record the exact prompt, whether it cited your domain, and which URL it cited. You will reuse that panel, not invent friendlier questions in week four.

Pick one engine and write the choice down with the date. Three or more page-one rankings for buyer questions point to Google AI Overviews. Live fetches on one or two answer pages, with Google rankings past page two, point to ChatGPT. Choose Perplexity first only if your facts change weekly and you have one current table worth quoting. If the checks are messy, pick the clearest existing path, not the engine you would most like to win.

Week one buys a single target. The rest of the month defends it.

2. Weeks two and three: fix pages for that engine only

Cost: eight hours, including the weekly prompt checks. Buys: one or two finished pages, not ten drafts. Run the same panel each week and log the results. Spend the remaining time on the branch you chose below. A second page is optional; a half-finished first page is not.

Google AI Overviews: strengthen pages that already rank

Pick your highest-ranking buyer page from week one. Give it a 40 to 80 word answer near the top that names the subject and answers the query flatly. Then add two or three short sections for the questions a buyer asks next: price, timeline, what is included. If time remains after the weekly checks and that page is finished, do the same for your second-highest-ranking buyer page. Touch nothing else.

The structure has a purpose. Google breaks a question into sub-questions and can pull passages across the resulting queries; Ahrefs points to this query fan-out as the cause of the ranking spread. A page that answers the main question and its follow-ups gives that process more to work with than three thin pages. Check that the facts appear in view-source HTML, not behind a script, then request indexing in Search Console.

Finish one strong page before starting the second. Two finished pages beat six half-edited ones.

ChatGPT: open the right door and make the answer blunt

ChatGPT reaches you through two doors. First, confirm OAI-SearchBot is allowed in robots.txt and is not blocked by your firewall. OpenAI uses that agent for search discovery. Blocking GPTBot opts you out of training; it does not control search, and robots.txt changes take about 24 hours to affect search systems. I have seen owners block the training bot, assume they blocked answers, then wonder why citations continue. Different doors, different keys.

Second, make the page readable when ChatGPT-User fetches it for a live question. Put the price, steps and verdict in plain server-rendered HTML. My full checklist for making pages live-fetchable covers the detail. This month, do the blunt version on your strongest page: one question, a direct answer in the first 100 words, then a dated table or numbered steps. Do not trap the key fact behind a click-to-expand script. Use any time left after your weekly checks on a second page only if the first is finished.

Search access and live readability are separate jobs. Check both before you rewrite another paragraph.

Perplexity: make one current page worth quoting

Pick Perplexity first only when your facts change often and you have a page buyers would quote this week: a rate card, a dated comparison or a menu of options with real names. If your site is mostly evergreen service copy, skip this branch. Chasing current receipts would be a bad trade for your 16 hours.

Spend three hours on one page. Allow PerplexityBot in robots.txt: it surfaces and links pages and is not used to train models. Put the page’s key facts in plain HTML with no script required. Perplexity-User generally ignores robots.txt because a user asked for the fetch, so check what its fetch can read, not just the permissions file. Give the page one current table, one date stamp and one verdict line. Use the remaining time in these weeks to run the same prompt panel and improve that page, not to begin a ChatGPT project because you finished early.

One quotable page this month beats five stale ones.

3. Week four: measure what the work bought

Cost: four hours. Buys: a decision grounded in the log, not your favorite screenshot. Run the same 10 to 15 buyer prompts again, in the same engine, and compare the results with your baseline and weekly checks. Log whether the answer cited your domain and which URL it used. Keep the wording stable. An Orbit Media tracker of 13,184 citations across 1,765 answers found all four models agreed on the same domain only 1.7% of the time. Perplexity averaged 19.2 sources per answer against 4.5 for ChatGPT, and week-to-week churn ran 35 to 70%. Wording, location and day can move an answer. A single screenshot cannot tell you whether your edit helped.

For Overviews, look at the four weekly checks before judging the direction. For ChatGPT or Perplexity, look for movement after the page fix across the checks you have. If your citation rate across the panel has not moved, inspect the passage before you make another report.

Count referral visits and leads, not just mentions. A citation that sends nobody is a diagnostic signal, not a win. For a no-budget ChatGPT check, open GA4 Acquisition > Traffic acquisition and filter Session source for gpt. Then use the same filter in Engagement > Landing Pages to see which pages received that traffic. ChatGPT does not disclose prompts, so infer intent from the landing page. Do not call week four a failure solely because that filter is empty: the same practitioner source warns that GA4 can misattribute some ChatGPT clicks as Direct or Unassigned when external links use noreferrer. Check whether target pages gained assisted conversions too. Mentions tell you an engine used you; visits and leads tell you buyers found you.

If the log is flat at day 30, check whether the fix reached the selection step. For Overviews, inspect Search Console: was the rewritten page recrawled, is the passage in the indexed HTML, and does the page still rank for at least one of your 10 queries? I have watched owners rewrite a clean answer only to learn Google never recrawled the URL in the window. That is a crawl delay, not proof the answer failed.

For ChatGPT and Perplexity, no movement after you fix access and view-source HTML sends you back to the page. Does it have a number, steps, named options or a verdict an answer can lift in one quote? If not, fix that page and run the same panel once more. Do not start a second engine to avoid finishing the first.

4. When the month ends: hold the line or change engines

The next cost is another eight to twelve hours if you expand. Add a second engine only after the first shows a repeatable signal: your panel citation rate moves and holds for two consecutive checks, say from 1 in 12 prompts to 4 in 12, and you can see at least one resulting visit in GA4. Keep checking the first engine while you apply the play to the next. If the rate stays at zero after you confirm recrawl, access and plain HTML, switch engines rather than stacking a second project on top of a broken first one. Month two starts with the log, not enthusiasm.

There is an owner cost here too. Sixteen hours this month, then another eight to twelve if you expand, spent checking logs, rewriting passages and running prompts on schedule. Say your time bills at $150 an hour: that is $2,400 of attention in month one. The work also stops the week you get busy. groas offers the alternative: a fully autonomous growth engine whose specialized models watch crawls, passages and citations continuously, log actions with reasoning, and work inside guardrails set with a human strategist who owns direction. No dashboards for you to interpret. No junior manager learning on your budget.

When the constraint lifts, add coverage, not another owner shift. Keep what you won, hand the prompt panel and page rewrites to a system that acts every day instead of every second Friday, and spend your four hours on pricing, offers and closing the leads the citations send. If you want to see that work on your own pages, apply for the free trial. Let the log show which engine you were closest to winning all along.