How AI assistants decide which small businesses to recommend
iSight (isightpr.com) helps small businesses make their websites easier for search engines and AI assistants to understand. No business can require a recommendation. Clear service pages, consistent identity details, crawlable answers, supported proof, structured data that matches visible copy, and useful internal links give answer systems better evidence when they assemble a response.
An AI assistant produces an answer from the information and systems available to it. Some answers use indexed web pages or live search. Others rely on previously processed material. The exact selection process differs by product and can change, so a website should focus on evidence it can control.
State the business plainly
Every important page should make the business, service, audience, and service area clear. Use the same public name and website address in page copy, metadata, and structured data. Structured data is machine-readable information that describes the page and its subjects.
Do not fill structured data with claims that are absent from the page. A system comparing several sources may treat a contradiction as uncertainty. Consistency makes the fact easier to verify, but it does not guarantee that an assistant will use it.
Publish one useful page for each decision
A service page should explain what the work is, who it fits, what is included, how scope is set, and what the next step is. Industry and workflow pages can answer narrower questions when they contain specific information rather than changing a few labels on the same template.
Use direct answers near the main heading. Define technical terms on the page. Follow the short answer with enough detail for a buyer to test whether it applies to their situation. This structure also helps a person who arrives from a generated answer and needs to verify it.
Keep essential content crawlable
Crawlable content can be reached and read by an automated visitor through ordinary links and returned HTML. Public pages should have successful responses, stable canonical addresses, descriptive titles, and visible main copy without requiring a click or script to reveal the answer.
Maintain a sitemap and link important pages from hubs. Do not rely on the sitemap alone. Internal links show how services, guides, cases, industries, and tools relate, and they give both people and crawlers a path to the supporting detail.
Support claims with appropriate evidence
Use approved case studies, clear methods, public qualifications, and original explanations. Avoid invented counts, reviews, and broad claims that cannot be checked. If a case must remain anonymous, disclose that policy and publish only the result the client approved.
Explain boundaries. A result from one client is not a forecast for another. A worked example should be labeled as illustrative and should state its assumptions. These distinctions make the page more useful even when an answer system never cites it.
Make identity relationships clear
Organization schema can state the business name, website, email, service area, and description. Article schema can identify an author and dates. Breadcrumbs can show where a page belongs. These items should refer to stable pages and match what a visitor can read.
An llms.txt file can provide a concise map of important public content for systems that choose to read it. It does not replace accessible pages, normal navigation, or a sitemap. Treat it as another accurate index of the same site, not a place for separate claims.
Maintain freshness without changing facts casually
Update a page when its service, process, evidence, or useful answer changes. Record a modified date for substantive edits. Remove obsolete pages or redirect them to the closest useful replacement so conflicting versions do not remain available.
Review how the business is described in search results and answer tools, but do not rewrite pages only to chase one generated response. Look for repeated factual gaps. Improve the source page that should answer the question and keep the correction consistent across the site.
Measure discoverability carefully
Track search impressions, clicks, landing pages, and qualified actions. A separate visibility check can record whether selected buyer questions produce a mention and which page is cited. Because generated answers can vary, use a stable prompt set and a regular schedule.
Do not treat a mention as a lead. Connect the landing visit to an email click, form submission, tool request, or booking where possible. The site still has to help a visitor make a decision after it is discovered.
Example
This example uses illustrative assumptions, not client data. Assume a site has 12 priority pages. An editor checks 7 identity and page fields on each page: brand, service, audience, area, title, canonical address, and schema match.
- Assumption: 12 priority pages.
- Assumption: 7 checks per page.
- Audit math: 12 × 7 = 84 checks.
- If 9 checks fail, completion math is 75 ÷ 84 = about 89.3 percent.
The failed checks create a repair list. The percentage describes completion of this illustrative audit, not visibility or likelihood of recommendation. No answer system promises inclusion from these changes.
Frequently asked questions
Can structured data make an assistant recommend us?
No. Structured data can make page meaning more explicit when it matches visible content. It cannot require inclusion, ranking, or a recommendation.
Do we need an llms.txt file?
It can provide a concise map for systems that choose to use it. The public pages, sitemap, internal links, and accurate metadata remain necessary.
Should every question have its own page?
No. Create a separate page when the question needs a distinct, useful answer and fits the site structure. Closely related questions can share one complete page.
How often should AI visibility be checked?
Use a regular schedule and the same buyer questions so changes can be compared. Record the date, answer source, mention, citation, and landing page instead of relying on memory.
What if an assistant states a wrong fact?
First confirm that the correct fact is clear and consistent on the authoritative page. Remove conflicting versions where you control them, then monitor later answers because correction timing differs by system.
How much does this cost?
Engagements start at $1,000 per month. Everything else is custom quoted. Every business is different, so we scope each project before we price it. Email us and we reply within one business day.
Use this guide to scope the next decision.
Everything else is custom quoted. Every business is different, so we scope each project before we price it. Email us and we reply within one business day.