Improve AI search visibility by making key pages crawlable, explaining your services with verifiable evidence, keeping business details consistent and testing the same buyer questions across platforms. Record whether each answer mentions or cites your business, whether it is accurate and whether relevant visits or enquiries follow. These checks measure observations; they cannot guarantee a recommendation.

Your business appears when people search its name but disappears when they ask AI tools for a provider in your category.

The cause could be site access, unclear service information or stronger evidence elsewhere. A single answer will not tell you which.

Use the 20-prompt baseline below to identify what to inspect before changing content.

What does AI search visibility mean?

AI search visibility describes whether systems such as ChatGPT, Google’s AI features, Gemini and Perplexity include a business or link to its pages when answering relevant questions. A mention names the business; a citation links to a source page. Record them separately.

Review four parts of the result:

  1. Access: Can the platform or its search partner reach the page?
  2. Clarity: Does the page explain the service, location, audience and supporting evidence?
  3. Appearance: Does the answer name the business, cite a page, do both or do neither?
  4. Business outcome: Do relevant visits or enquiries follow?

An answer that mentions a company does not show whether the description is accurate or whether a buyer visited. Track those outcomes separately.

How do you build a 20-prompt baseline?

Use a repeatable set of questions that reflect how buyers look for help. Do not rewrite a website based on one favourable or unfavourable answer.

The following examples cover five useful question types. Replace them with the services, locations and decisions that matter to your business. For a comparable baseline, run all 20 prompts on each platform or mode you choose to check and keep each set separate. If time is limited, start with one platform or mode.

Use five groups with four prompts each.

Category prompts

These test broad discovery.

  • Which web design companies serve small businesses in Pretoria?
  • Which process safety consultants operate in South Africa?
  • Which firms offer corporate landscaping in Mpumalanga?
  • Which South African agencies help businesses improve AI search visibility?

Problem prompts

These start with a business need.

  • Who can redesign a service website that receives traffic but few enquiries?
  • Which company can help a factory prepare for a HAZOP study?
  • Who can build a professional website for a company without an internal marketing team?
  • Which provider can help a South African business become easier to find in ChatGPT and Perplexity?

Comparison prompts

These ask an engine to evaluate choices.

  • Compare three Pretoria web designers for a five-page business website.
  • Compare a freelancer, small studio and agency for a South African SME website.
  • Which South African AI visibility services publish their method and limitations?
  • Compare once-off website pricing with monthly website packages in South Africa.

Evidence prompts

These test whether public proof exists.

  • Which South African web designers publish detailed pricing and ownership terms?
  • Which agencies show case studies that explain the business problem and result?
  • Which local providers explain how they measure ChatGPT and Perplexity recommendations?
  • Which firms publish original South African research in their service category?

Branded accuracy prompts

These test entity understanding after the engine receives the name.

  • What services does IDJOY provide?
  • Where is IDJOY based?
  • Does IDJOY publish web design prices?
  • What evidence supports IDJOY’s AI search visibility service?

Discovery and branded prompts answer different questions. Keep them separate in the report.

What should you record for each test?

For each prompt, save the conditions and result needed to repeat the test:

Field What to record
Date and time Exact run time
Engine ChatGPT, Gemini, Perplexity or Google AI feature
Product or mode Search, deep research, AI Mode or another named mode
Context Stated location and whether you were logged in
Exact prompt Unedited wording
Result Whether the business was mentioned and a website page cited
Other sources URLs or source types cited in the answer
Accuracy Correct, incomplete or wrong, with the specific issue

Run each prompt in a fresh conversation and record the product or mode. Earlier conversation context can affect the answer. Record sentiment, competitors or next actions only when they change your decision.

A baseline records answers under stated conditions. It does not produce a universal ranking.

Can the platforms access your site?

Crawler access is one part of the check. A platform may still mention a business from other sources even when its crawler cannot access the site.

Google

Google’s AI features guidance says pages must be indexed and eligible to appear with a Search snippet to be considered as supporting links in AI Overviews or AI Mode. Google does not guarantee that an eligible page will be crawled, indexed or shown.

Google says there are no additional technical requirements or special schema types for these features. Its generative AI search guide also says llms.txt and other special AI files are not needed for Google Search.

ChatGPT

OpenAI’s ChatGPT Search guidance says sites should allow OAI-SearchBot if they want their pages to appear in search answers. Allowing the crawler makes pages eligible; it does not guarantee a mention or citation.

OAI-SearchBot supports search discovery. GPTBot is a separate control for potential model training. Check the user-agent rules you need before changing robots.txt.

Perplexity

Perplexity’s crawler documentation says PerplexityBot helps surface and link websites in Perplexity search results. Perplexity recommends allowing that crawler and its published IP ranges if you want the site to appear there.

Perplexity separately documents Perplexity-User, which may fetch a page in response to a user’s question and generally ignores robots.txt because the user initiated the request.

What to inspect

Check:

  • robots.txt rules.
  • noindex tags.
  • Canonical URLs.
  • Firewall and bot-protection rules.
  • Pages that require JavaScript before the main answer appears.
  • Redirect loops.
  • Duplicate query-string and clean URLs.
  • XML sitemap coverage.
  • Server logs where available.

When should you create a separate page?

Google’s generative AI search guide advises against making a separate page for every query variation to influence results. Google says a high page count alone does not make a site higher quality or more relevant.

Group similar phrases when they need the same answer. Create another page only when it answers a distinct question or decision.

For example, these queries belong to one pricing pillar:

  • web design prices South Africa;
  • web design South Africa pricing;
  • web design rates South Africa;
  • web design cost South Africa;
  • website design costs South Africa.

If several pages answer the same pricing question, combine shared information on the primary pricing page. Give a separate page to a distinct decision, such as first-year costs, quote comparison or freelancer-versus-agency models.

What evidence makes a page specific to your business?

First-hand details can help a page stand apart from general advice. Google’s guide recommends useful, original content, while making no promise that it will appear in an AI feature.

Useful original evidence includes:

  • A real project decision and why it changed.
  • An anonymised scope document.
  • A current price or starting range with exclusions.
  • A test method and raw prompt list.
  • A before-and-after information structure.
  • An industry checklist based on field work.
  • A client-approved result with dates and limits.
  • A documented failure and the correction.
  • A local regulatory or operational explanation.

Do not manufacture statistics to look authoritative. A small honest sample beats an invented percentage.

Use evidence that a reader can inspect. Do not describe a prompt test as a result unless you can provide its prompts, conditions and recorded answers.

Which business facts should you keep consistent?

Compare the facts on your website with the public profiles and sources that buyers or answer engines may encounter. Inconsistent details can lead to an inaccurate description.

Review:

  • Legal and trading name.
  • Primary domain.
  • Phone number.
  • Address or service area.
  • Services.
  • Industries served.
  • Founder and team names.
  • Opening hours.
  • Social profile links.
  • Pricing claims.
  • Old brand names and domains.

When an answer cites an outdated listing, correct that listing if you control it. Adding another page does not change the details on a third-party source.

What should a visitor find on the service page?

When a reader follows a citation, the destination page should answer:

  • Who the service is for.
  • What problem it solves.
  • What is included.
  • What is excluded.
  • How the work happens.
  • What evidence supports the offer.
  • What it costs or how pricing works.
  • What the visitor should do next.

If you are comparing paid support, check the deliverables, exclusions, measurement method and price.

Which independent sources can confirm your business?

A company cannot establish every claim through its own website.

Third-party sources can confirm:

  • Business existence.
  • Location.
  • Reviews.
  • Professional memberships.
  • Industry participation.
  • Published expertise.
  • Partnerships.
  • Project work.

Focus on legitimate sources that serve the audience. Do not buy hundreds of low-quality directory links or fake mentions.

A useful external source should help a person verify the business even when no AI system existed.

Which structure helps readers find the answer?

Clear headings, direct answers, tables, definitions and descriptive links help people scan complex pages. Google says special chunking is not required for its Search features; use structure that makes the subject easier to follow.

Use structure because the subject needs it:

  • Put the direct answer near the top.
  • Give tables clear column names.
  • Keep one main question per section.
  • Link claims to primary sources.
  • Name dates and locations.
  • Use descriptive internal links.
  • Keep important content in crawlable HTML.

What can Google Search Console measure?

Google’s Search Console Help says the generative AI performance report for AI Overviews and AI Mode rolled out worldwide on 31 August 2026. A property may not show the report if it has too few qualifying AI-feature impressions. The report measures Google Search; it does not report ChatGPT, Gemini or Perplexity answers.

Check the correct property and date range before concluding that a South African site is excluded. The report groups impressions by pages, countries, devices and dates. Google’s announcement says the data is also included in the overall Search performance report.

Combine it with:

  • Standard Search Console performance.
  • ChatGPT referrals, when your analytics identify them; record them separately from manual prompt tests.
  • Other identifiable AI referrers.
  • Server-log crawler activity.
  • Form submissions and call tracking.
  • Lead-source questions.
  • Manual prompt observations.

Search Console reports Google Search. Your analytics and lead records are needed to connect other identifiable referrals with enquiries and sales.

When should you rerun the prompt baseline?

Choose an interval you can keep consistent. Retest after meaningful changes to your site, service details or measurement setup, and record the interval with each run.

Reuse the same core prompts so you can compare observations. Record new questions in a separate group to keep the baseline intact.

For each platform or mode, calculate mention share as the number of its 20 prompts that name the business divided by 20. Calculate citation share as the number of those prompts that cite a business page divided by 20. When you repeat the same 20 prompts, calculate mention and citation test volatility separately: count how many prompts changed status since the previous run, then divide by 20. These are observations for your recorded conditions, not universal rankings.

Track referrals, engaged visits, enquiries and sales separately when your analytics and lead records can identify them. Do not treat one mention or referral as proof of a sale.

Which tactics do not guarantee AI visibility?

Adding llms.txt

Google says it does not use llms.txt or other special AI files for Search. Check each platform’s own documentation before treating an optional file as useful there.

Treat it as an optional documentation convenience, not an AI visibility strategy.

Adding schema without matching page content

Structured data can clarify content. It cannot create services, proof, prices or locations that the visible page does not explain.

Publishing many near-identical AI articles

Google warns that publishing pages for query variations to manipulate Search results can violate its scaled content abuse policy. A large number of pages does not make a site more useful by itself.

Repeating “best” claims

A self-declared leadership claim does not give the reader comparison criteria or independent evidence.

Paying for guaranteed organic recommendations

Google says meeting its requirements does not guarantee that a page will be crawled, indexed or shown. Treat a promise of guaranteed organic recommendations across AI platforms with caution and ask the provider to define what it can actually control.

What sequence should you follow?

Establish a baseline and check access

  • Run the 20-prompt test.
  • Check crawlers, indexing, canonicals and sitemap coverage.
  • Map the sources engines cite.
  • Correct inconsistent business facts.
  • Prioritise questions tied to a service purchase where answers are inaccurate, incomplete or do not mention the business.

Improve the source pages

  • Strengthen the main service pages.
  • Consolidate duplicate content.
  • Add supporting content only when it answers a distinct buyer question and you have evidence to support it.
  • Add case studies, scope examples or clear pricing where possible.
  • Improve internal links from articles to services and proof.

Retest and connect results to outcomes

  • Rerun the fixed prompts.
  • Compare mentions, citations and accuracy.
  • Review Google and AI referral traffic.
  • Inspect which pages attract engaged visits.
  • Record qualified enquiries and assisted decisions.
  • Choose the next improvement from the evidence.

If the crawler is blocked, fix access first. If pages are accessible but answers omit or misstate important facts, improve the relevant service pages and public business information. If answers are accurate but do not lead to enquiries, review referral and lead data before publishing more content. Repeat the same prompts under the same conditions before choosing the next change.

How does IDJoy approach AI visibility?

IDJoy works with South African service businesses that want to connect their website, content and measurement.

The process includes:

  • A fixed prompt baseline.
  • Technical access checks.
  • Service-page improvements.
  • Evidence-led content.
  • Internal linking.
  • Business-fact consistency.
  • Quarterly prompt retesting in the tracking scope.
  • Reports that separate observations from analytics.

IDJoy’s tracking scope repeats 20 questions across eight engines each quarter. Review the AI search visibility service for current pricing and deliverables.

IDJoy does not guarantee recommendations, first positions or a fixed result date.

The service is designed for businesses that can contribute first-hand information and want a reviewable process. It does not include submissions that buy organic placement or bulk content production.

For measurement detail, read the AI recommendation measurement guide before choosing a provider.