AI Search Visibility
AI Search Visibility Software: First-Week Setup Guide
A day-by-day first-week setup for AI search visibility software: prompts, competitors, engines, baselines, and a 3-day Noma trial test.

To set up software for AI search visibility, configure prompts first, then competitors, engines and a baseline measured over several days. Published research says single runs are unreliable, so use the first week to build a stable measurement setup. Noma's 3-day free trial suits a setup check, with billing starting on day 3.
To set up software for AI search visibility, configure four things in the first week, in this order: a prompt library written from buyer questions, a short competitor list, the AI engines your buyers use, and a baseline measured over several days rather than one run. Do this before you read any dashboard trend, because a tool with the wrong inputs produces confident-looking noise.
This guide is for the buyer who has just chosen a platform, or is about to start a trial, and wants a day-by-day plan. It also shows how the 3-day free trial of Noma (a card is taken at signup, billing starts on day 3, and nothing is charged if you cancel before then) fits a setup test. If you are still choosing, start with the comparison of AI search visibility tools first.
Key takeaways
- Write 20 to 30 prompts from real buyer questions, grouped by intent, before you judge any dashboard. The prompt library is the measurement instrument.
- Pick three to five direct competitors and track the same prompts for all of them from day one, so share of voice has a denominator.
- Prioritise engines by where your buyers ask, not by the tool's longest list. Noma tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily, so you decide which ones to read first.
- Treat any single run as an anecdote. Published research says one query on a given day can differ substantially from a second one minutes later, so baseline over days and weeks.
- Use a 3-day trial to validate setup and data quality, not to judge trends. Decide on day 3, before billing starts.
- Fix crawler access and brand facts in the same week, because they decide whether the engines can read you at all.
Why does the first week of AI visibility software matter?
Most tools in this category look similar in a demo: a dashboard, a list of engines, a share-of-voice chart. The difference between a useful and a useless deployment is almost always the inputs. Tool roundups from Guideflow, SE Visible and Exposure Ninja mostly compare features and starting prices. They say little about what to do after the purchase, which is where this guide sits.
The week-one decisions are hard to reverse in one respect: your baseline is only comparable with later data if the prompts, engines and competitors stay stable. Change them every few days and you lose the ability to say whether a content change worked. That is why the order of setup matters more than the speed of it.
AI answers are probabilistic. A paper on measuring visibility in AI search, posted to arXiv in April 2026, concludes that visibility is unstable and cannot be treated like traditional SEO rankings. Setup should be designed around repeated measurement.
That paper, "Don't Measure Once" (arXiv, April 2026), states that practitioners should use at least 7 runs per prompt per day for brand visibility monitoring, and at least 8 runs when source-level coverage matters. It studied ChatGPT, Perplexity, Gemini and Google AI Mode, with up to 10 runs per engine and prompt. We add one practice of our own: keep baselines separate per engine, because the engines behave differently.
What should I configure first in an AI visibility tool?
Configure prompts first, competitors second, engines third and baselines fourth. Prompts come first because every other setting is applied to them: competitors are scored against your prompts, engines answer your prompts, and a baseline is a measurement of your prompts. A tool with 13 engines and a poor prompt list measures the wrong thing 13 times.
| Order | Setting | What it decides | Cost of getting it wrong |
|---|---|---|---|
| 1 | Prompt library | Which buyer questions you are measured on | Every metric reflects the wrong questions |
| 2 | Competitor set | The denominator for share of voice | Share of voice is inflated or meaningless |
| 3 | Engines and markets | Where you focus your reading of answers | You miss the engine your buyers use, or pay attention to ones they do not |
| 4 | Baseline window | The reference point for every later change | You cannot tell signal from noise |
| 5 | Access and brand facts | Whether engines can read and describe you | Fixes you ship later have no effect |
How do I set up AI search visibility software, step by step?
The six steps below are a generic process that works with any tool in the category. They take about a week of part-time effort. Each follows the same pattern: the action, the result you should have by the end, and why it matters.
Step 1: Build the prompt library from buyer questions
Action: Write 20 to 30 prompts in plain conversational language, sourced from sales calls, support tickets, search queries and the questions your own FAQ answers. Group them into four intents: discovery ("what is the best option for X"), comparison ("A vs B"), problem ("how do I fix Y") and brand ("is [your brand] good for Z"). Result: a spreadsheet of prompts, each tagged with an intent, a funnel stage and a product line.
Why: HubSpot's guidance calls prompt tracking the core unit of measurement in this category and suggests moving from awareness-stage prompts to decision-stage ones, per its AI search analytics tools guide (HubSpot, checked November 2026). A practical 15-day plan from Omnia (checked November 2026) also starts with 20 to 30 prompts across three to four clusters. Our own advice on which prompts to track and how many prompts you need goes deeper on selection.
| Intent | Example pattern | What it tells you | Suggested share of library |
|---|---|---|---|
| Discovery | "Best [category] for [use case]" | Whether you are recommended unprompted | Largest group |
| Comparison | "[Brand] vs [competitor]" | How you are framed against rivals | Second largest |
| Problem | "How do I [task your product solves]" | Whether your content is cited as a source | Medium |
| Brand | "Is [brand] worth it" | What engines say when asked directly | Small, but check every week |
The suggested shares are our own working heuristic, not a published benchmark. Adjust them to your funnel. Avoid prompts that contain your brand name in the discovery group, because they test recognition rather than recommendation.
Step 2: Define a competitor set of three to five
Action: List three to five direct competitors, including at least one that you suspect is already cited more often than you. Add the category names buyers use, since engines often answer with a list of brands. Result: a fixed roster applied to every prompt. Why: HubSpot's baseline workflow runs the same prompt library against the top three competitors and logs their baseline scores, and Omnia recommends three to five. Without a roster, a rise in your mentions cannot be read as a change in position.
Keep the roster stable for at least a month. If a new rival appears in the answers, record it in a watch list rather than swapping it in mid-baseline. Our guide to AI search share of voice explains the calculation that depends on this list.
Step 3: Prioritise engines and markets by where buyers ask
Action: Rank engines by where your buyers are likely to ask category questions, then read those first. Note country or language if you sell in more than one market. Result: a short list of priority engines, with the rest reviewed at lower attention. Why: engines differ, so a blended number hides the difference between them. Keep a separate baseline for each engine.
| Engine group | Typical buyer behaviour | Setup note |
|---|---|---|
| ChatGPT | Conversational research, shopping and comparison questions | Check OAI-SearchBot access (see Step 5) |
| Google AI Overviews and AI Mode | Questions asked inside Google Search | Same indexing and snippet rules as Search |
| Perplexity | Source-heavy research with visible citations | Allow PerplexityBot in robots.txt |
| Gemini and Claude | Assistant-style research and writing help | Track as separate baselines, not blended |
Noma tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily. In a first week, read the engines in the table first, then scan the rest to see whether any surprising engine already cites you. If you sell mostly in one region, see our notes on AI search visibility tools in Germany for how market settings change what you check.
Step 4: Set a baseline over days, not one run
Action: Collect answers for several consecutive days before changing anything on your site. Record, per prompt and per engine, whether the brand is named, which competitors appear, and which pages are cited. Result: a baseline with a visibility rate per engine and a list of your cited pages. Why: one answer proves little. Cloro (checked November 2026) reports that about 300 answers per period resolve a visibility rate to roughly ±5 percentage points, and that a claimed move from 20% to 30% needs about 294 answers on each side.
Worked example. Suppose you track 20 prompts on 3 engines. That is 60 prompt-engine pairs. Collect each pair once a day for 5 days and you have 300 answers, matching the cloro figure for a ±5-point estimate. The arithmetic is ours; the 300 and ±5 come from the source. A 3-day window gives 180 answers, enough to catch a gross problem such as zero mentions on an engine, but not enough to claim a 10-point gain.
One caution: the 1-run-a-day rate in this example is for illustration only. It is well below the arXiv paper's recommended minimum of 7 runs per prompt per day, so treat the figures as a floor for reasoning about sample size, not as a target. A tool that polls more often than once a day will reach these counts sooner.
| Window | Answers (20 prompts, 3 engines, 1 run a day) | What you can conclude |
|---|---|---|
| 1 day | 60 | Rough picture only |
| 3 days | 180 | Gross gaps, setup errors, who the main competitors are |
| 5 days | 300 | A starting rate of roughly ±5 points, per the cited source |
| 14 to 28 days | 840 to 1,680 at the same rate (our arithmetic; weekdays only would give 600 to 1,200) | A steadier baseline; we suggest two to four weeks of rolling aggregation |
For deeper guidance on checking a vendor's figures, read how to verify AI search tool data. If answers differ by user, personalization in LLM responses explains why.
Step 5: Open crawler access and check the pages engines read
Action: Check that your robots.txt, CDN and firewall allow the search crawlers of the engines you track. Confirm that key pages are indexable and that the facts about your brand (name, category, location, product list) are the same everywhere. Result: a short access log with allowed or blocked for each crawler. Why: a tool can only measure outcomes; it cannot make a blocked page citable.
OpenAI's help centre, in its guidance on allowing OpenAI web crawlers (OpenAI, checked November 2026), recommends allowing OAI-SearchBot, and warns that a 403 on an image host can block product feed processing. Perplexity's crawler documentation (Perplexity, checked November 2026) recommends allowing PerplexityBot in robots.txt and permitting its published IP ranges. Google states in AI features and your website (Google Search Central, checked November 2026) that sites appearing in AI Overviews and AI Mode are included in overall Search Console traffic, and that snippet and noindex controls govern what is shown.
Use our AI crawler user agents list to decide what to allow, and the entity consistency checklist to align brand facts. The GEO readiness scorecard covers both in one pass.
Step 6: Turn the baseline into alerts and a first action list
Action: Choose three alert conditions, for example a competitor replacing you in a top discovery prompt, a drop to zero mentions on an engine, and a new third-party source cited for your category. Then list the five most fixable gaps. Result: an action list ranked by effort and likely effect. Why: a baseline that does not lead to a task is a report, not a programme.
Our guides on AI search monitoring alerts and the AI visibility action plan show how to turn tracked data into tasks. When you change a page, follow the method in GEO experiment design so that you can attribute the result.
Which first-week checklist should I follow?
Use this ordered list as the single page your team ticks off. It assumes one owner and about two hours a day.
- Collect 40 raw buyer questions from sales, support and search data, then cut to 20 to 30.
- Tag each prompt with intent, funnel stage and product line.
- Remove brand-name prompts from the discovery group.
- Name three to five competitors and freeze the list for a month.
- Rank engines by buyer use and read the top group first.
- Note country and language if you sell in more than one market.
- Confirm crawler access for OAI-SearchBot and PerplexityBot, and check Google snippet controls.
- Check that brand name, category and key facts match across your site and major profiles.
- Review the tool's answers daily for at least three days and note any prompt that fails to return an answer.
- Record per-engine visibility, competitor presence and cited pages.
- Set three alerts and write the top five fixable gaps.
- Decide whether to keep the tool, using the pass or fail criteria in the next section.
How does a 3-day free trial fit a setup test?
Noma offers a 3-day free trial. A card is taken at signup, billing starts on day 3, and you are not charged if you cancel before then. Three days is too short to see a trend, so the right use is to prove that the setup works and the data is trustworthy, then decide. This is the same logic as the AI visibility tool trial test plan.
| Day | Task | Pass condition |
|---|---|---|
| Day 1 | Sign up with your prompt library, competitor roster and priority engines agreed on paper; see how your brand and category appear in the dashboard | Your brand and main competitors show up across the engines you care about |
| Day 2 | Review the first polled answers; spot-check five against the engine by hand | Recorded mentions match what you see manually |
| Day 3 | Read the citation sources, competitor comparison and recommendations; cancel or continue | You can name three actions the data supports, before billing starts |
Because Noma polls daily, three days gives you roughly three observations per prompt per engine, if every prompt is polled each day. That is a setup check, not a baseline. If the pass conditions hold, continue past day 3 and keep your baseline running for two to four weeks before drawing conclusions. If you cancel, do it before billing starts on day 3. Ask these 10 questions for an AI search visibility demo if you want a vendor to walk through anything the trial left open.
Noma by Nurdd is an AI search visibility dashboard that tracks whether a brand appears in answers from 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily. It records whether a brand is named and which page each citation comes from, compares competitors and recommends actions. That makes it suitable for the baseline, competitor and engine steps in this guide.
Want to see this on your own brand? Book a demo and we will show how your brand and competitors appear across the 13 engines in Noma, and what a first-week baseline looks like for your category.
What should the baseline scorecard record?
A baseline is only useful if you record the same fields every time. Keep one row per prompt and engine, then summarise by engine and by intent. Do not blend engines into a single number until you have seen the per-engine values side by side.
| Field | Definition | Why it earns a column |
|---|---|---|
| Named | The brand appears in the answer text (yes or no) | The basic visibility rate |
| Position | First, listed, or passing mention | Being named first differs from being one of ten |
| Competitors present | Which rostered rivals appear | Feeds share of voice |
| Cited page | The URL the engine used for your brand, if any | Shows which of your pages work as sources |
| Third-party sources | Non-owned sites cited in the answer | Shows where to earn mentions beyond your site |
| Framing | How the answer describes you | Catches wrong or outdated descriptions |
The cited-page and third-party columns are the most commonly skipped, yet they point to the actions in Step 6. Our guide to tracking an AI search visibility score week over week shows how to keep the scorecard consistent once the baseline closes.
How do I check whether the tool is giving me real data?
Run a three-point audit. First, pick five prompts and ask each engine yourself; the mentions in the tool should broadly match, allowing for the variation the research describes. Second, open the cited sources and confirm the pages exist and say what the tool reports. Third, compare the same prompt on two consecutive days and note how much the cited sources change. Large variation is normal; a tool that shows none is worth questioning.
| Check | Method | Red flag |
|---|---|---|
| Answer match | Compare five prompts by hand | Tool reports mentions you cannot reproduce at all |
| Source match | Open each cited URL | Citations link to pages that do not mention the claim |
| Stability | Compare consecutive days | Identical output every day, or no history of runs |
| Method transparency | Ask how answers are collected and how often | No explanation of sampling |
Our monthly AI search metrics report template shows how to turn the checked numbers into a report.
What are the most common first-week mistakes?
- Starting with the tool's suggested prompts only. Suggestions are useful, but they come from the tool, not your buyers. Mix both.
- Tracking too few prompts. One or two prompts reflect the quirks of those prompts. Spread coverage across intents.
- Blending all engines into one score. Keep engine-specific baselines.
- Changing the site before the baseline closes. You lose the reference point.
- Reading day-over-day swings as trends. Cloro's sample-size figures show a single day is too small a sample to compare.
- Ignoring access. A blocked crawler cannot be fixed by a dashboard.
- Skipping the owner. Name one person who reads the alerts weekly.

How does Noma help after the first week?
After setup, Noma shows citation frequency by category, a competitor comparison and recommended actions to improve AI visibility, so the baseline you built becomes a working list. Because it polls 13 engines daily, the two-to-four-week baseline builds without manual runs. If you want the team to carry out the fixes, Nurdd's generative engine optimization services and answer engine optimization agency work shape the content and sources engines read, measured in Noma. For how the products fit together, see how Nurdd works.
If you are weighing rivals, the Noma vs Profound comparison and the guide to GEO, AEO and SEO explain the field. For a longer plan after week one, use the 90-day AI search marketing plan.
How we researched this
We reviewed platform documentation from Google, OpenAI and Perplexity, one preprint on measuring visibility in AI search (checked against its HTML version), a sample-size analysis from an AI visibility vendor, setup guidance from HubSpot and Omnia, and three tool roundups that currently rank for this topic. All pages were checked on the publication date of this post. We did not run hands-on tests of the tools named here, and the example figures marked as our arithmetic are illustrations, not measurements. Vendor pages describe vendors' own claims.
Frequently asked questions
How do I set up AI search visibility software?
Write a prompt library from real buyer questions, add three to five competitors, prioritise the engines your buyers use, and build a baseline over several days before changing your site. Then check crawler access and set a few alerts. The six steps above take about a week.
What should I configure first in an AI visibility tool?
Prompts. Competitors, engines and baselines are all applied to your prompt list, so errors there spread through every metric. Aim for 20 to 30 prompts grouped by intent.
How many prompts do I need to start?
A common starting range is 20 to 30 across three to four clusters, according to Omnia's published plan. You can add more once the baseline is stable. See our guide on how many prompts to track for the trade-offs.
How long until I can trust the data?
One run is not reliable. The arXiv paper recommends at least 7 runs per prompt per day for brand visibility monitoring. We suggest rolling aggregation over two to four weeks. A few days will show gross gaps, but not small changes.
Can I test AI visibility software in three days?
Yes, for setup and data quality, not trends. Noma's 3-day free trial takes a card at signup and bills on day 3 unless you cancel first, so use day 3 to decide.
Which engines should I track first?
Start with those your buyers use for category questions, usually ChatGPT, Google AI Overviews, Perplexity and Gemini, then Claude. Keep baselines separate per engine.
Do I need to change my website in week one?
Only to fix access and factual consistency. Hold content changes until the baseline closes, so you can attribute later movement to what you changed.
Is manual checking enough?
For a handful of prompts, yes. At 20 to 30 prompts across several engines and days, software saves hours and keeps a record. Our comparison of manual prompts and software covers where the line sits.
The first week decides whether your AI visibility data will be usable. Build the prompt library, fix the competitor set, pick engines deliberately, and hold off on conclusions until the baseline has days behind it. Book a demo to see how your own brand and competitors appear across the 13 engines in Noma, or start the 3-day free trial and run the day-by-day plan above.