You have 60 days, one owner, 20 buyer questions, and no paid tools. Spend the first two months watching rankings and you will miss the earlier signal: whether an assistant came to read the answer you wrote.

I used to tell clients to watch rankings while we built content. I was wrong. Rankings became a poorer guide once AI answers started sitting above the links. In one Ahrefs study of 863,000 search pages, only 37.9% of AI Overview citations ranked in the top 10, down from about 76% the prior year. The same study found AI Mode and AI Overviews cited the same URLs only 13.7% of the time. A position report cannot tell you which answer an assistant chose to read.

So here is the trade: stop tracking rankings for 60 days. Write one answer page for each question buyers actually ask an assistant. Then check server logs for live fetches from ChatGPT-User and Perplexity-User. A fetch is not a citation, let alone a sale. It is the earliest observable sign in this plan that an assistant came to read a page. For Google AI Overviews, you will check citations by hand; those fetch logs do not give you the whole picture.

The budget is about 40 hours total: roughly 10 hours a week for four weeks of hands-on work, spread across the first 50 days, with the final stretch mostly for watching and checking. No site-wide rebuild, no 100-keyword project, no visibility-score subscription. The plan is to clear the barriers to being read, publish 20 useful answers, earn a few outside mentions, and see what gets picked up. Here is where every hour goes.

1. Days 1–5: Choose the 20 questions (6 hours)

Cost: 6 hours. Buys you: a writing list tied to buying decisions. Start with 10 questions from the last 90 days of sales calls, quote requests, and support tickets. Keep the buyer’s wording. Then find 10 more by asking ChatGPT and Google AI Mode what someone in your trade would ask before hiring you. Try your service plus your city; inspect the follow-up questions they suggest.

Keep questions with money behind them: price, timeline, guarantees, comparisons, and failure cases. Cut a question if its answer would not change who gets hired. “Tips for choosing Y” can wait. “How much does X cost?” cannot, assuming you can answer it honestly.

Run each survivor through one more filter: can you give a direct answer in about 40 words, including a useful number? The point is not to staple a statistic onto every sentence. It is to make the answer specific enough to stand on its own. Cited passages often cluster around 30 to 60 words, with the answer appearing in the first 40 words of a section. By the end of day 5, you want 20 questions written as buyers ask them. Those phrases become your page headings.

Cut keyword-volume reports, difficulty scores, and competitor-gap exports. On this budget, they can consume two writing days without giving an assistant a better answer to quote. Do save one small piece of reconnaissance for later: search your brand plus your service and read the first five pages. Note the directories and lists that appear. Ask ChatGPT the same thing and note the sites it cites. Put both columns in a plain sheet. That is your outside-mention list for days 36–50, not a second research project.

Leave this step with 20 buyer questions, not 20 keyword targets.

2. Days 6–10: Make the pages readable to assistants (5 hours)

Cost: 5 hours. Buys you: a site that does not block the work you are about to publish. Assistants cannot quote an answer they cannot access. Check yoursite.com/robots.txt in plain text. Do not disallow OAI-SearchBot for ChatGPT Search, Claude-SearchBot for Claude, or PerplexityBot for Perplexity. Blocking GPTBot or CCBot concerns training crawls; allowing those bots alone is not a plan for appearing in answers. Check the search crawlers that matter to the pages you want found.

Then check the floor, not the wallpaper:

  1. Confirm your sitemap is linked.
  2. Make sure the answer is visible in plain HTML, without a login wall.
  3. Check mobile loading against the 2.5-second target.
  4. If JavaScript is the only way to see the answer, put the answer where a plain page fetch can read it.

That is the technical sprint. No schema detour and no wholesale site redesign.

Finally, learn where to look for requests. In your server logs or Cloudflare, search for ChatGPT-User, Perplexity-User, OAI-SearchBot, and PerplexityBot. ChatGPT-User and Perplexity-User are the live-answer fetches to watch; search-crawler hits are a different signal. Check that your logs actually capture the requests you intend to count. Once the new pages go live, repeat the search weekly. If they record no relevant fetches after 30 days, check access before spending another hour polishing prose.

If an assistant cannot fetch the answer, better writing will not fix the access problem.

3. Days 11–35: Publish one answer page per question (20 hours)

Cost: 20 hours, about one hour per page. Buys you: 20 answers an assistant can lift without assembling a puzzle. One question gets one page. Do not bury all 20 in a single FAQ. Open each page with the question as its H1, then answer it in the first two sentences, in under 40 words. Follow with a specific, relevant number and its named source and year. Explain the mechanism after that.

This is a format choice with a reason behind it. In Princeton tests on 10,000 queries, adding quotations, statistics, and citations raised visibility 30 to 40% on a position-adjusted word-count measure. An answer that includes its own context is easier to use than a passage beginning “as noted above.” That does not mean every page with a number will be cited. It means you are giving the page a better shot at being useful.

I would write each page in the same order: answer, number, source, mechanism, boundary. Suppose you are a garage door company. An illustrative opening might give the installed price range for a standard double garage door replacement in Austin and say it takes one day. The next sentence must say where that range came from and when. Then explain what changes the quote: labor, door weight, opener reuse. Finish with who should skip that option. The boundary helps a buyer decide, and it gives an assistant something more useful to repeat than a sales pitch.

Do not turn the illustrative price into a made-up local fact. If you cannot support a range, do not print one. Pick a question you can answer from your own documented pricing or a named source instead. The same rule applies to timelines and guarantees: an honest limit is more useful than a neat figure you cannot stand behind.

One hour per page is a hard cap, not permission to publish filler. Use the buyer’s exact question, make the opening do the work, and spend the remaining time on the explanation a buyer needs before calling. Skip rewrites of old blog posts. Skip the introductory paragraph that announces what the page is “going to cover.” The buyer asked a question. Answer it.

If one week collapses, protect the 20-page goal by cutting optional polish, not by replacing the plan with a smaller library. Keep the direct opening, the supported number, and the explanation. Leave decorative formatting and old-post renovations for day 61. A published answer can be fetched; the perfect draft in your documents cannot.

4. Days 36–50: Get your facts repeated elsewhere (6 hours)

Cost: 6 hours. Buys you: outside corroboration your own pages cannot provide. An assistant has more than your description of your business to draw on. In Ahrefs data from 2026, branded web mentions correlated 0.664 with AI Overview appearances, versus 0.218 for backlinks. Brands were 6.5 times more likely to be cited through third-party sources than through their own domains. Those figures do not promise you a citation for fixing a listing. They tell you why outside pages deserve hours in a 40-hour plan.

Think of it as your blog versus other people’s pages, with different jobs for each. Your page carries the answer. Their page helps establish that the business behind it exists and describes itself consistently.

Return to the list from day 5. Spend roughly four hours claiming or correcting the most relevant profiles: Google Business Profile, Yelp, G2 or Capterra if you are software, Trustpilot or BBB if you are local. Do not create five bad profiles just to hit five. Keep your name, address, phone number, and service facts identical where you have listings. Businesses with active review profiles appeared in 75.3% of AI answers versus 1% for those with none; listings on G2, Capterra, or Trustpilot were also associated with higher citation probability.

Use the remaining two hours to pitch three local or trade sites for a factual mention: a supportable price range, service area, or year founded. You are not buying links or sending a “thought leadership” essay nobody requested. You are making a few verifiable facts easier to find beyond your own site.

Cut mass directory submissions. A handful of accurate profiles beats a stack of thin, inconsistent ones.

5. Days 51–60: Read the scoreboard you can actually use (3 hours)

Cost: 3 hours. Buys you: a decision about which pages to keep improving. Do not open a rankings report. Open your server logs. Filter requests for ChatGPT-User, Perplexity-User, OAI-SearchBot, and PerplexityBot; group them by answer-page URL and sort. Keep live-answer fetches separate from search-crawler visits. Verified live-answer fetches are more useful here than raw AI-bot hit counts. A live fetch shows that an assistant came for the page. It does not show that the answer used it.

Now check that next step by hand, once a week. Ask ChatGPT and Google AI Mode your 20 questions in plain buyer language, in a fresh session. Note which pages, if any, they name. Keep two counts in a sheet: live fetches per page and citations per question. Add a note when an inquiry mentions an assistant. You are not building an attribution empire in three hours; you are separating read, cited, and contacted.

Use the differences to decide what to edit:

  • Fetches but no citations: the assistant can read the page. Tighten the opening 40 words, check the source and number, and make the answer stand alone.
  • Citations but no inquiry: keep watching. A named page is not yet evidence that a buyer chose you.
  • No fetches after 30 days live: recheck access, then reconsider the question or whether the answer says anything distinct. Do not apply this verdict to a page published late in the sprint that has not had 30 days.

Google AI Overviews still need the manual check; these user-agent counts are not a complete record of what Google considered. That limitation is a reason to label the columns correctly, not a reason to go back to rankings. Edit a page that gets read but not cited. Reconsider one that is accessible but never fetched.

6. Day 61: Lift the constraint only if the log earns it

Cost against the 60-day budget: zero hours. Buys you: a rule for what happens next. Keep doing it yourself if the pages have not shown repeat live fetches or the citation checks give you nothing to build on. Revisit the questions and the answers before paying for more output. Volume is not the cure for an answer nobody reads.

If three or more pages show repeat ChatGPT-User fetches and one cited answer produces an inquiry, the limit may be your pace instead. You cannot hold the work to 10 hours a week and also publish daily, clear technical blocks as they appear, and pursue outside mentions while running the business. That is when a done-for-you service becomes a reasonable next expense, not an act of faith.

Groas runs SEO & AI Search for a flat $1,499 a month, with new pages and technical fixes every day and no cut of ad spend. Keep your 20 pages and the log habit as the baseline. Let the added capacity build on the answers buyers and assistants have already shown they will use. Do not pay for volume until the fetch log gives you a reason to.