Skip to content

GEO & AEO for Feature Previews

Use GEO (generative engine optimization) and AEO (answer engine optimization) to promote features you haven't built yet. Turns real user questions into honest feature preview pages with design mockups and a design review log, written so Google AI Overviews, ChatGPT search, Perplexity, Copilot and Claude can find, quote and cite them, then tracks daily traffic, AI referrals and waitlist sign-ups to decide build / rework / drop (with a script).

Intermediate⏱ 2-3 h setup + 10 min/day📂 launchv1.0.0

In an afternoon you will have 1–3 honest preview pages for features you're designing, set up so search and AI engines can cite them, plus a daily log that tells you after 28 days which feature to build.

Goal

Find out which features people want before you build them, by publishing one preview page per feature (problem, mockups, design decisions, waitlist) that answers real questions well enough to be found in Google and quoted by AI assistants. The finished output is: the pages, a tracking setup, and a build / rework / drop call per feature backed by your own traffic and sign-up numbers.

New to GEO and AEO? Read the learning guide first (40 min). It covers the terms, how AI answers pick sources, what Google, OpenAI, Perplexity, Anthropic and Bing officially say, the GEO research, crawler setup and how to read daily numbers.

When to use

  • You have mockups or a design review, no code yet, and want demand evidence before building.
  • You want the product to show up in AI answers ("best app for…", "how do I…") from day one.
  • You have a site (even a one-page one) where you can publish pages and edit robots.txt.
  • Not for: a live product that needs store installs → aso-keyword-research. Finding the questions in the first place → research-where-users-ask. A full launch checklist → launch-pre-launch-checklist.

Inputs

  • 1–3 features you are designing, each with at least one mockup (Figma export, screenshot of a prototype).
  • Notes from your design review: decisions, what you cut, open questions.
  • 10+ real questions or complaints about the problem (threads, reviews, autocomplete). No questions → do research-where-users-ask first.
  • Access to your site, Google Search Console, Bing Webmaster Tools and GA4 (all free).
  • A waitlist form (email-list-and-deliverability if you have none).

Steps

  1. Set up measurement before publishing (30 min). This is day 0 of your log; without a baseline you can't tell growth from noise.
  2. Verify the site in Google Search Console and Bing Webmaster Tools (Bing has an AI Performance report of Copilot citations).
  3. In GA4, add a custom channel "AI assistants" above Referral (regex in the guide, section 8), and a key event on the waitlist form with the feature name.
  4. Copy assets/daily-traffic-log.csv (the rows in it are illustrative; delete them) and log the last 7 days as baseline.
  5. Open the door to crawlers (10 min). robots.txt allows Googlebot, Bingbot, OAI-SearchBot, PerplexityBot, Claude-SearchBot; training bots (GPTBot, ClaudeBot, Google-Extended) are your choice. Check your host's bot protection doesn't block them. Pages must be in the sitemap and not noindex.
  6. Pick the question per feature (20 min). From your collected questions, choose the one main question per feature and 4–6 sub-questions (pricing, platform, "vs X", "does it…"). These are your H1 and H2s. Write them in the users' words, not your feature name. Before writing, ask ChatGPT (search on), Perplexity and Google the main question and note who is cited: that's your competition and your outreach list.
  7. Write one preview page per feature (40 min each). Use the template: status line ("In design · not available yet"), a 40–60 word short answer, question H2s with the answer first, labelled mockups, a "vs. today" table with cited facts, the design decisions from your review, what it won't do, an FAQ, and one action (waitlist or vote). Add owned facts: a number from a prototype test, a quote from your designer or a tester (real, with permission).
  8. Run the honesty check (5 min). No Download / Buy / Trial buttons, no reviews or user counts you don't have, every image labelled Design mockup, no near-duplicate pages per keyword. These break Google's spam policies (misleading functionality, doorway, scaled content) and the FTC reviews rule; see the guide, section 6.
  9. Publish and get indexed (15 min). Link each page from the home page or a /roadmap page, submit the sitemap, request indexing in Search Console (URL Inspection) and Bing (URL Submission). Write the publish date in the log.
  10. Build presence in 3 places in week 1 (60 min). AI engines cite other sites too. Answer 3–5 real threads where the question lives (link only where it helps, disclose it's yours), post a 30-second mockup walk-through with captions, and list the product in 1–2 directories that already show up for the question. UTM-tag every link.
  11. Log daily, read weekly (10 min/day, 15 min/week). Each day add visits by page and source plus sign-ups. Each week run the prompt check (5 prompts × 3 runs × ChatGPT, Perplexity, Google; record "mentioned 1/3") and: 
    python3 scripts/traffic_trend.py daily-traffic-log.csv --start 2026-10-01
    It compares the latest 7 days with the 7-day baseline, splits search / AI / social / direct, shows the 7-day rolling average and sign-ups per 100 visits per page. Change one thing per page per week and log the date.
  12. Decide at day 28. Per feature: build (traffic growing and people sign up), rework the page (flat traffic: the question or short answer is off), rethink the feature (traffic but no sign-ups), or drop. The script's thresholds (3 visits/day minimum, +10% vs baseline, 2 sign-ups per 100 visits) are this repo's starting rules of thumb, not industry benchmarks: tune them to your niche. Email the waitlist the decision.

Prompt (copy-paste)

Works in Claude, ChatGPT, Gemini, Cursor. Replace everything in {{ }}. Attach mockups if the tool accepts images.

You are a GEO/AEO editor for a small software team. We are promoting a feature we have NOT built yet.
Write an honest feature preview page that search engines and AI assistants (Google AI Overviews, ChatGPT search,
Perplexity, Copilot) can find, quote and cite.

Product: {{product, platform, who it's for}}
Feature: {{name + one line}}
Status: in design, target {{month year}}
Real user questions (verbatim, with links where public): {{10+ questions}}
Design review notes: {{decisions, what we cut, open questions}}
Mockups: {{list what each mockup shows}}
Facts we own: {{prototype test numbers, quotes with names and permission; or "none yet"}}
What people use today: {{competitors / workarounds, with source links}}
Waitlist URL: {{url}}

Do this:
1. Pick the ONE main question for the H1 and 4-6 sub-questions for H2s, in the users' words. Explain the choice in 2 lines.
2. Write the page in Markdown:
   - Status line at the top: "In design · not available yet", target month, last updated date, author.
   - "The short answer": 40-60 words that make sense on their own.
   - Each H2 is a question; the first sentence under it answers it directly.
   - A comparison table vs. what people use today, with every outside fact linked to its source.
   - "Design decisions so far" from the review notes, and "What it won't do (yet)".
   - FAQ (availability, price, platform), then one call to action: join the waitlist or vote.
   - Mark every image "Design mockup".
3. Rules: no Download/Buy/Trial buttons, no invented reviews, numbers, users or logos; if a fact is missing, write
   {{TODO: fact}} instead of guessing. No keyword stuffing. Write for a person first.
4. After the page, list: the title tag (under 60 chars), meta description (~150 chars), 5 prompts I should check
   weekly in AI assistants, and 3 places (threads, directories) where I should answer this question this week.

Example output

Illustrative example. Fictional product and made-up numbers to show the format.

Snapsort, a Mac photo organizer in design. Feature: batch rename. Page /features/batch-rename:

Status: In design · not available yet. Planned for Feb 2027. Last updated 1 Oct 2026 by Linh, product designer.

The short answer. Batch Rename will rename hundreds of photos on a Mac by date, place or camera in one step. Unlike Finder's rename, it shows every new file name before applying it and can undo the whole batch.

Can I rename photos by the date they were taken on Mac? Yes, that's the default pattern…

Daily log, day 28, from traffic_trend.py:

Page Visits (last 7 d) vs. baseline AI visits Sign-ups / 100 visits Call
/features/batch-rename 73 +480% 7 5.5 Build: keep designing, interview 5 waitlist people
/features/smart-albums 41 +270% 5 4.9 Keep the page; test a sharper short answer
/features/cloud-sync 9 +12% 0 0.0 Rethink: people don't search for this

Prompt check, week 4: "best app to batch rename photos on Mac", Perplexity: mentioned 1/3 (cited page: /features/batch-rename); ChatGPT 0/3 (cites two Reddit threads → answer those).

Common mistakes

  • "Download now" on a feature that doesn't exist. It's misleading functionality to Google and to people. The only action is waitlist, vote or feedback.
  • Judging after 3 days. New pages take days to weeks to be indexed and longer to be cited. Read the 7-day average and decide at day 28.
  • Changing five things at once. One change per page per week, dated in the log, or you'll never know what worked.
  • Writing feature-name H2s ("Introducing SmartSort™"). Nobody asks that. Use the questions people type.
  • Only your own site. AI answers lean on Reddit, reviews and directories. Without presence, you get read but not recommended.
  • Ignoring sign-ups. Traffic says people look; sign-ups per 100 visits say they want it. Build on the second.
  • Trusting the AI-visits number as total. Some assistants send no referrer; that traffic hides in Direct.

Credits

Written by @nvminhtu. Sources: Google Search Central (AI features and your website, spam policies), OpenAI crawlers, Perplexity crawlers, Anthropic crawlers, Bing AI Performance, Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024), the FTC reviews rule, and the MIT-licensed ai-seo skill in coreyhaines31/marketingskills. Checked 18 Oct 2026; re-check the official pages each quarter.