If you built a website by prompting an AI, its navigation probably uses the words you gave the builder. Your customers may search with very different words. To find SEO keywords for a vibe-coded website, start with the problem the visitor is trying to solve, collect plausible queries for your country and language, inspect the current results, and assign each useful query group to a page that can actually answer it. Do not turn every variation into a new page.
This guide walks through that decision with an explicitly fictional appointment-reminder website for independent salons. It includes a completed query-to-page map, including ideas to reject, so you can repeat the method for your own site. The site-building tool is less important than the public pages and the value they deliver. If you still need the broader sequence of indexing, audits and page creation, start with the vibe coding SEO guide. Here we focus on one task: deciding which search terms deserve your next page or page update.
Translate your website into customer problems
Write down what someone wants before they know your product exists. An AI builder prompt might say “create a dashboard with automated CRM workflows.” A salon owner may ask how to remind clients about tomorrow's appointments or reduce missed bookings. The second language is where useful seed topics begin.
Describe the site's real offer in one sentence. Then list three to five problems it can credibly help with. For our fictional example, suppose the website offers appointment reminders and a client confirmation flow. Possible problem seeds are “remind clients about appointments,” “reduce salon no-shows,” and “confirm appointments online.” “Best salon booking software” would be a different promise if this product does not provide booking. Reject it rather than stretching the page to capture an attractive term.
Collect wording from customer questions, support tickets, sales calls and reviews when you have permission to use them. A new website may have none. In that case, ask a few people in the target audience how they describe the task, then compare the phrases with actual search results. Treat an AI-generated list as brainstorming, not evidence that people use those phrases. Avoid inserting the name of your builder into customer-facing keywords unless your offering really serves builders.
Record each seed with the intended visitor and the solution your site can show. This early constraint protects the project from publishing unrelated articles for every term a research tool suggests. Google's people-first content guidance asks whether the content serves an existing or intended audience and leaves readers able to achieve their goal. For a more detailed page-writing brief after you choose a topic, use Screpy's SEO content brief guide.
Gather query candidates for one market
Pick the country and language your website serves before comparing keyword ideas. “Appointment reminders” in one market may have different demand, phrasing and competitors from another. If you serve more than one language, build separate lists and check whether your site has genuinely localized pages before planning them.
Start with each problem seed and collect variants from autocomplete suggestions, related searches, customer vocabulary and the titles of current result pages. Note the wording, source, market and date in a simple sheet. Search suggestions are clues, not a complete measure of volume. Check the live search results for each promising candidate; they reveal what users seem to expect and whether the results are service pages, tutorials, templates or tools.
Screpy's Keyword Research can expand a seed for a selected country and language and show estimated volume, difficulty, intent and trend signals. Use those numbers to narrow the review, not to predict clicks. The research guide also recommends validating results against the current SERP and any available Search Console pages. A “low difficulty” label alone does not show that your specific website can satisfy the query.
If the site is already receiving impressions, Google's Search Console Performance report can show which queries led to appearances and which pages received them. Filter by query and page, then ask whether the right page is appearing. On a new or little-seen site, that table can be thin; absence of a query there is not proof that no one searches it. Screpy's Search Console guide explains how to inspect connected first-party data inside the product.
Keep a short candidate set. Ten plausible terms that you can examine carefully are more useful than a thousand suggestions you cannot map to actual site content. For each, write a one-line question: “What would the searcher want to do after clicking?” That question leads to the next filter.
Inspect the results before deciding what to build
Search the candidate phrase in the target market and look at the result types, not just the words in their titles. If the first page mainly shows step-by-step guides, a thin product feature page may disappoint that visitor. If it shows pricing and product comparisons, a generic “what is” article may answer an earlier question but miss the buying decision. Search results change and can vary by location and device, so record what you saw and when.
Classify each query by the task behind it: learn a process, compare choices, evaluate a service, or complete an action. These categories are useful working labels, not a universal Google taxonomy. Open several representative pages. What question does each answer first? What proof or functionality does it provide? Could your site offer something genuinely useful beyond rephrasing the same advice? This manual check matters more than a tool assigning an “informational” or “transactional” badge.
Watch for mixed results. “Salon appointment reminders” could mean message templates, software, or a general how-to, depending on the market. Do not decide from the phrase alone. If intent is unclear, keep the candidate under review and test a more specific variant. A modifier such as “software,” “text examples,” or “how to” can expose a clearer task. But a longer phrase is not automatically less competitive or more valuable.
Check your existing URLs before deciding to make another one. If /salon-reminders/ already answers the query, improve that page's clarity and evidence first. Creating a near-copy at /salon-appointment-reminder-software/ can split the same intent across two weak pages. Conversely, a useful instructional article may deserve its own URL when the service page cannot answer the question without burying the buying information. The decision is about what the visitor needs, not about assigning one keyword to every possible route.
Turn the shortlist into a query-to-page map
Now give every candidate an outcome: improve an existing page, group it with a close variant, create a genuinely different page, investigate further, or reject it. Do not let the keyword tool decide the site architecture. The following fictional map assumes a salon appointment-reminder website with a homepage, a salon use-case page, a pricing page and a blog. The phrases are examples, not measured search volumes or findings from a live SERP.
| Candidate query | Likely task to check in live results | Page decision | Why |
|---|---|---|---|
salon appointment reminder software |
Evaluate a solution | Improve /salon-reminders/ |
The existing use-case page should explain the actual reminder flow, show a realistic example and link to pricing. |
salon reminder software |
Possibly the same evaluation task | Group with /salon-reminders/ if results confirm |
A wording variant alone does not justify a second page; keep the terminology aligned with the actual website. |
salon appointment reminder text examples |
Find usable message examples | Consider a new examples page | Only publish if you can provide real, permission-safe examples, context and constraints beyond a generic list. |
how to reduce salon no shows |
Learn a process | Improve an existing guide or create one if absent | A guide can cover booking confirmation, timing, communication and measurement without pretending software alone solves every cause. |
salon booking software |
Book/manage appointments | Reject for this site | Reminder software without a booking capability cannot honestly answer the query. |
best appointment reminder software |
Compare vendors | Hold for evidence review | A credible comparison requires current alternatives, transparent criteria and accurate claims; a self-serving list is not enough. |
The table's “likely task” column is a hypothesis. Search the query in your own market and change the decision if the results show another intent. This makes the map a working document rather than an SEO template to publish unchanged.
For every approved query group, write one primary page URL, the visitor question, the proof you can provide and the action the visitor can take. Keep a note of queries you rejected; they are useful guardrails when an AI coding assistant later suggests more pages. If you decide that a dedicated use-case page is justified, the next task is to write and validate it. The separate guide on SEO landing pages for vibe-coded websites covers that build step.
Google's spam policies distinguish helpful pages from content produced mainly to manipulate rankings at scale. A route list created by swapping “salon,” “barber” and “spa” into otherwise identical AI copy does not become useful merely because each URL has a unique title. Require meaningful, page-specific value before adding a route.
Choose the first page by fit, not one metric
After the map, you still need an order of work. Choose one page update or new page that passes four checks: the query fits the real offer, the visitor's task is clear, your site can add specific value, and the change is feasible to maintain. Use estimated demand and difficulty to inform that choice, but do not turn them into a magic score. A low-volume query can be useful if it brings the right visitor; a high-volume query can be a poor target if it asks for a feature you do not have.
For the fictional salon site, improving /salon-reminders/ may be the first task because it already exists and can answer a product-evaluation query more honestly. A message-examples page is a second candidate only if the team can write useful examples and review them for real-world suitability. The “best software” page remains on hold until the team can do a fair, current comparison. This is an editorial priority, not a prediction of ranking or conversion.
Use a small worksheet with columns for query group, target URL, visitor question, intended action, evidence you can provide, maintenance owner and next step. The maintenance owner matters: a comparison that goes stale or an examples page with obsolete guidance can damage trust. Mark “update,” “build,” “hold” or “reject.” Keep the rejected ideas visible so the next builder session does not recreate them by accident.
When you ask an AI assistant to implement the plan, supply the reviewed worksheet and specify the one route to change. Ask it to show how the page answers the visitor question and where the internal link belongs. Then inspect the published output yourself. The assistant can implement a page; it cannot establish demand, product truth or visitor satisfaction from its own generated text.
Revisit the map using real search data
Once the page is live and accessible, record the publication or update date and verify its public URL. Search Console's query and page views can later tell you whether Google has shown the intended page for relevant searches. Look at impressions, clicks and the actual landing URL together. A rise in impressions without useful visits is a reason to inspect the query and snippet, not automatically to produce more articles. On a young website, the data may be too sparse for confident conclusions, so keep customer feedback and direct visitor questions alongside it.
If a query appears against the wrong page, check whether the intended page exists, is discoverable, answers the task and links sensibly from related pages. If two pages keep competing for what is really the same visitor need, improve or consolidate rather than creating a third. If a new query reveals a distinct problem that the current page cannot answer without becoming muddled, add it to the research queue and repeat the SERP check. Screpy's connected Search Console view can help inspect query and page data; it does not replace the judgment about whether a new page would be useful.
Keep the map small and current. Keyword research for a vibe-coded website is not a one-time list pasted into an AI prompt. It is a set of evidence-backed page decisions: which visitor questions you can answer, where those answers belong, and which attractive terms you should leave alone. That gives the next build session a clear brief and gives your visitors fewer, stronger pages to use.