AI4SMB.agency
Get Started
All posts

AI Search and SEO: How Small Businesses Can Be Found and Represented Accurately

AI-narrated audio

A prospective customer asks an AI search tool to recommend a local supplier. Your business appears in the answer—but the opening hours are wrong, the price comes from an old PDF and the link points to a page that no longer exists.

Being mentioned is not the same as being understood. For a small or mid-sized business, the practical challenge is to make useful information findable, keep it consistent across the places customers consult, and give people a reliable way to enquire or buy.

Traditional search engine optimisation (SEO) still matters. AI-assisted search adds ways for information to be retrieved, combined and presented before customers reach your website. A practical response starts with accurate core information, accessible pages and a working customer journey.

The immediate priorities

If time is limited, start with five things:

  1. Check that your most important pages can be accessed and indexed.
  2. Correct conflicting details about your services, prices, availability and opening hours.
  3. Maintain relevant business profiles, including Google Business Profile and Bing Places for Business.
  4. Make it easy for customers to verify your offer and take the next step.
  5. Measure accurate representation, enquiries and sales alongside search visibility.

These priorities support three distinct outcomes: discovery, representation and action.

What AI search changes

Search results already contain more than links: maps, product listings and direct answers can all shape a customer's decision. Generative AI adds responses that synthesise information from multiple sources, often with supporting links and conversational follow-up questions. Different services may use web retrieval, model knowledge or a combination of both.

Google says AI Overviews and AI Mode can run several related searches to assemble a response. To qualify as a supporting link, a page must be indexed and eligible for a search snippet. Meeting those requirements does not ensure selection. 1

You may hear answer engine optimisation (AEO), generative engine optimisation (GEO) or AI SEO. Google treats this work as SEO because its generative search relies on existing ranking and quality systems. Other providers have their own rules. 2

For business owners, the more useful distinction is:

  • Discovery: Can a relevant service find and access your public information?
  • Representation: Does its answer describe your business accurately, for the right customer and market?
  • Action: Can an interested person verify the details, contact you, book or buy?
A citation is evidence of an appearance. It does not establish that the whole answer is correct or that a customer took action.

An answer might also influence a purchase without producing a visit your analytics can identify.

What the traffic research tells us

A 2026 research preprint analysed browsing behaviour from 1,100 US Google users in a randomised experiment. Participants experienced ordinary Google Search, a version attempting to suppress AI features, or searches redirected into AI Mode. Assignment to AI Mode reduced external click-through rate by 18.8 percentage points relative to ordinary Search; the main analysis counted each AI Mode prompt as a search. This is a percentage-point difference, not an 18.8% fall in every website's traffic. 3

The treatment lasted seven days and forced AI Mode use rather than observing voluntary adoption. Participants came from Prolific and a university work-study programme, with an unrepresentative demographic profile. Google's interface changes also broke some AI Overview suppression in the separate comparison condition. These limitations restrict generalisation to long-term behaviour and European small businesses. 3

Google's own account is different in scope. In August 2025, it reported relatively stable aggregate organic clicks year on year and more clicks where users did not quickly return to Search, which it called higher-quality clicks. That is a dated, company-reported aggregate—not a conversion measure or evidence about an individual site's performance in 2026. 4

The two accounts measure different things over different periods. Neither tells a particular business what will happen to its enquiries.

Use them as a reason to investigate your own customer journeys. A definition may be answered entirely in search. A complex purchase may still require specifications, pricing, reassurance and a conversation. Your exposure depends partly on which needs your website serves and how those visits contribute to revenue.

Start with access

Before commissioning a content programme, inspect a small set of important pages: your homepage, priority service or product pages, contact page and one useful resource.

Check indexing and the live page separately

In Google Search Console, use URL Inspection to review the indexed version and run a live test. Look for access errors, unexpected redirects, an unintended noindex instruction and missing content in the rendered page.

The indexed report describes what Google has processed. The live test checks the current page's accessibility and potential indexability; it does not confirm that the latest version is indexed. Review the indexed report separately for Google's selected canonical URL—the version it treats as the main page. 5

If important content is inaccessible, investigate hosting, firewall, content delivery network and crawler settings before rewriting it. A sitemap can assist URL discovery, but submission does not guarantee crawling or indexing. 6

When a page moves, redirect its old address to the relevant replacement. Avoid sending unrelated old pages to the homepage: Google warns that irrelevant redirects can confuse visitors and be treated as errors. 14

Review search access and AI controls

A crawler visits pages on behalf of a search or AI service. The robots.txt file gives instructions to cooperating crawlers. It cannot secure confidential material; use appropriate authentication and access controls for that. 13

For Google, check Googlebot access and review Settings → Search generative AI in Search Console. Google's documented control covers AI Overviews, AI Mode and generative features in Discover, with worldwide rollout completed by 31 August 2026. A property may inherit its setting from a parent property. Exclusion concerns those features; Google says it is not a ranking or inclusion signal for other parts of Search. Training controls are separate. 7

OpenAI likewise distinguishes search and training crawlers. OAI-SearchBot supports ChatGPT search; GPTBot concerns content that may be used to train foundation models. A site can allow search crawling while disallowing training crawling. OpenAI says sites opting out of OAI-SearchBot are excluded from ChatGPT search answers, although navigational links may still appear. User-requested visits through ChatGPT-User are a separate mechanism. 8

Review these choices with whoever manages your website and content rights. Permission to crawl makes content available for consideration; it does not buy a recommendation.

Make the offer clear

A technically accessible website can still leave customers guessing. “Innovative solutions for a changing world” gives little help to someone deciding whether you can solve a specific problem.

A priority page should answer:

  • What do you provide, and who is it for?
  • Where is it available, and in which languages?
  • What is included, excluded or subject to confirmation?
  • What do prices mean: which currency, market and conditions apply?
  • What evidence supports your claims?
  • What should the customer do next?

Use a descriptive title and main heading, readable text and useful links to supporting information. Keep essential details available as text alongside images, video or interactive elements. Google's AI-search guidance supports these familiar SEO practices. 1

Publish knowledge customers can use

Choose topics where your business can contribute something specific: a compatibility limit, a comparison based on experience, a realistic implementation process or an explanation of a common purchasing mistake. Google's broader content guidance emphasises original information, demonstrable experience and clear sourcing. 15

For example, a technical distributor could explain which specifications matter for a particular installation and link to current manufacturer documentation. A hotel could describe its actual late-arrival procedure and identify requests requiring staff confirmation. These pages help customers make decisions and give intermediaries clearer material to work with.

Create separate pages when they answer distinct needs. A directory of near-identical city pages is difficult to maintain and offers little value if the underlying service is unchanged. Google warns against producing variations primarily to manipulate search and says there is no ideal length for AI-search content. 2

Use genuine testimonials, qualifications and references. Make it possible to check them. Commissioning large volumes of generic copy will not resolve an unclear offer or an unsupported claim.

Treat conflicting information as an operational problem

Consider a fictional specialist retailer. Its product page says delivery covers Spain and Portugal, an old catalogue says Spain only, and its FAQ shows a discontinued delivery charge. A customer or assistant could encounter three different versions of the offer.

First establish the correct answer with the team responsible. Then update or retire conflicting sources. If a historical PDF must remain available, label it clearly and link to the current information. Check translated pages, product feeds, marketplaces and profiles under your control as well.

An information register makes this work repeatable. For each high-impact fact, record:

  1. Fact and scope: The opening hours, delivery region, price or policy, including the market and conditions.
  2. Current reference: The page or operational system maintained as the authoritative business record.
  3. Other appearances: FAQs, catalogues, listings, feeds and translations that repeat it.
  4. Owner and trigger: Who approves changes and what prompts a review, such as a seasonal schedule or product withdrawal.
  5. Correction status: What has been fixed, what depends on a third party and when it will be checked again.
This register is an internal process, not a special search signal. Its value is reducing contradictory information and making updates someone's responsibility.

When an AI answer contains a consequential error, save the answer and cited sources. Correct the material you control and request changes to relevant third-party pages. Check again over time: editing your website does not establish when a provider will reprocess it or whether another source will continue to repeat the mistake.

Maintain local, language and product information

Local business profiles

Check your business name, contact details, opening hours, service area and public address where applicable. A business visiting customers should explain where it operates and whether customers can visit its premises.

Maintain Google Business Profile and, for relevant retailers, Google Merchant Center. Google explicitly recommends keeping both current. 1

Add Bing Places for Business to the review. Microsoft recommends accurate address, hours and contact information there to support eligibility for local AI responses. This complements the website and is particularly relevant when monitoring Bing and Microsoft Copilot. 12

Languages and regional offers

For businesses serving multiple European markets, review the offer in each important language. Check service boundaries, local contact routes, delivery conditions, currencies and tax presentation. Translating a headline while leaving contradictory conditions elsewhere can create expensive misunderstandings.

Google recommends separate URLs for language versions and supports hreflang annotations to identify language and regional alternatives. This helps Google understand those relationships; it does not maintain translations or determine what every AI service will select. 9

Product details and feeds

For ecommerce, compare names, identifiers, variants, prices, stock and shipping conditions across visible pages and the feeds you provide. Assign responsibility for keeping them aligned. An integration that updates less reliably than your shop introduces another source of stale information.

Before investing in a new feed, confirm the provider's current eligibility rules, supported markets and maintenance requirements. Start with the channels your customers use and that your team can keep accurate.

Use structured data where it helps

Structured data is machine-readable information describing page content. It can support eligible search features when implemented correctly and kept consistent with what visitors see. Google requires accurate, relevant markup and does not guarantee a rich result even when the technical implementation is valid. 10

Google also states that its generative search features require no special schema or structured data. It says llms.txt and similar special AI files do not affect Google Search visibility. Other services may adopt different mechanisms; verify their documentation before adding another source to maintain. 2

Evaluate an optimisation proposal by what it will change and how success will be assessed. Ask which pages, profiles or customer journeys need work, which platforms can be measured, and whether the supplier will check factual accuracy. Treat guaranteed recommendations or citations as claims requiring strong evidence.

Support images and usable customer journeys

Relevant product photographs, diagrams and demonstration videos can help customers understand an offer. Google says its generative search can surface images and video as well as page links. Keep descriptions, captions and visible product details consistent with the accompanying text. 2

Some AI agents also interact with websites to perform tasks. Google's web.dev guidance recommends predictable interfaces, accessible structure and clear feedback to help these interactions. 16

For a small business, a useful first step is to test the ordinary customer journey: can someone understand the form, choose a valid option, see the total cost and tell whether a booking or enquiry succeeded? Those improvements have immediate value for human customers while making the site easier for automated systems to interpret.

Measure visibility, accuracy and business outcomes separately

Record a baseline before making substantial changes. For a small site, the previous 28 days of available search and enquiry data can be a manageable starting point. Note priority pages, markets, promotions and major site changes. Compare equivalent periods later, allowing for seasonality, delays and small numbers.

A short before-and-after comparison can reveal a useful pattern. It rarely isolates the effect of a particular edit.

Understand Google's two reporting views

Search Console's Generative AI performance report (Search) reports link impressions in AI Overviews and AI Mode. Google says worldwide rollout was complete by 31 August 2026; the report may be absent when a site has insufficient impressions. It supports page, country, date and device breakdowns. It does not provide a dedicated breakdown of AI clicks, conversions or the exact prompts triggering appearances. Discover has a separate report. 11

Activity from these Search features is also included in the overall Search results Performance report under Web. That broader view combines search activity, so it cannot by itself establish which enquiries came from AI answers. 1

Do not add the dedicated AI impressions to overall impressions as though they were separate audiences. Differences in aggregation can also make chart totals and page totals differ. 11

Interpret Bing's citation reporting carefully

Microsoft introduced AI Performance in Bing Webmaster Tools in public preview on 10 February 2026. It reports citations across Microsoft Copilot, Bing AI summaries and selected partner experiences, including cited pages and trends. Citation counts do not indicate rank or placement. Its grounding-query phrases are a sample of retrieval activity, not a complete list of customer prompts. Check current availability and coverage. 12

Connect the evidence to customer outcomes

Use several measures together:

  • Visibility: Which priority pages appear in supported reports, and how does that change?
  • Accuracy: Are the location, offer, conditions and contact details correct in sampled answers?
  • Website outcomes: Which visits lead to useful enquiries, bookings or purchases?
  • Customer evidence: What do customers say influenced their decision?

Check how analytics classifies referrals and expect gaps in attribution. An optional “How did you hear about us?” question can add context, although recall is imperfect. Configure analytics and consent processes appropriately for your markets and tools.

Run a repeatable manual check

Choose a small set of real customer questions: local discovery, a service comparison, a product or policy question, and a brand-specific accuracy check. Include important languages and both questions naming your business and those that do not.

Use the services relevant to your customers. For each test, record:

  • Date, service and model or version if visible.
  • Exact question and follow-up questions.
  • Language and location settings.
  • Whether web search was enabled.
  • Sign-in state and whether the conversation was fresh.
  • Whether the business was mentioned or linked, what was correct or wrong, and which sources were cited.

Repeat on different occasions. This is a diagnostic sample, not a representative ranking or market-share estimate. Wording, location, account context and product changes can affect answers.

One favourable screenshot does not establish consistent visibility. An omission does not establish a technical failure.

When an answer is wrong, investigate its sources before deciding what to change.

A practical first-month plan

Assign one coordinator and involve the people responsible for the website, commercial information and customer enquiries.

Week 1: Establish the baseline

Select three to five priority pages and a handful of buyer questions. Save available search data, record meaningful enquiries and run initial manual checks. Identify access or indexing problems.

Week 2: Correct high-impact inconsistencies

Review the facts most likely to lose a sale or disappoint a customer: availability, price conditions, opening hours and contact routes. Update the current reference pages, Google Business Profile, Bing Places and relevant feeds or listings. Start the information register.

Week 3: Improve one customer journey

Make one priority product or service page clear and specific. Answer the main buying questions, provide evidence and test the enquiry, booking or checkout process. Add a page only where a distinct customer need justifies it.

Week 4: Recheck and establish ownership

Repeat the page and answer checks. Record unresolved third-party errors and assign review triggers for future changes. Continue monitoring beyond the first month; search processing and buying cycles may take longer.

Keep the scope proportionate. If search contributes little to your business, maintaining accurate core pages and reliable contact routes may justify more attention than producing a large library of articles. If an important offer is inaccessible or misleading, address that first.

Build a business that is easy to verify

The fictional retailer cannot dictate every AI answer. It can establish a current delivery policy, correct conflicting catalogues and make the offer easy to confirm. That improves the information available to customers, search systems and assistants alike.

AI search changes how people discover a business and how that business is described before a customer arrives. The practical goal is to be findable, represented accurately and easy to deal with. Measure each of those outcomes—and keep responsibility for the underlying information inside the business.

Sources and further reading

All links were checked on 19 November 2026. Platform documentation changes; consult the linked pages for current controls and reporting scope. The numbered links in the article point directly to the corresponding sources below.

  1. Google Search Central. AI features and your website. Eligibility, technical foundations and inclusion in overall Search reporting. For newer controls and dedicated reports, see sources 7 and 11.
  2. Google Search Central. Optimizing your website for generative AI features on Google Search. SEO, content, multimedia and unsupported optimisation shortcuts.
  3. Wang, Stephanie T.; Gleason, Jeffrey; Bart, Yakov; Wilson, Christo; and Metaxa, Danaé. AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence. arXiv:2608.18352v1, 18 August 2026. Research preprint; see methods, limitations and supplementary Table S8.
  4. Reid, Liz, Google. AI in Search is driving more queries and higher quality clicks. 6 August 2025. Google's own aggregate traffic account.
  5. Google Search Console Help. URL Inspection tool. Indexed information, live testing and canonical URLs.
  6. Google Search Central. Learn about sitemaps. Discovery benefits and indexing limitations.
  7. Google Search Console Help. Search generative AI control. Included features, exclusion, inheritance and rollout status.
  8. OpenAI. Overview of OpenAI Crawlers. OAI-SearchBot, GPTBot and ChatGPT-User distinctions.
  9. Google Search Central. Managing multi-regional and multilingual sites. Separate URLs, language selection and regional annotations.
  10. Google Search Central. General structured data guidelines. Accuracy, visible content and rich-result eligibility.
  11. Google Search Console Help. Generative AI performance report (Search). Impressions, dimensions, aggregation and coverage.
  12. Madhavan, Krishna; Merchant, Meenaz; Canel, Fabrice; and Nigam, Saral, Microsoft Bing. Introducing AI Performance in Bing Webmaster Tools Public Preview. 10 February 2026. Citation metrics, sampled grounding queries and Bing Places guidance.
  13. Google Search Central. Introduction to robots.txt. Crawler instructions and their security limitations.
  14. Google Search Central. How to move a site. URL mapping, redirects and avoiding irrelevant destinations.
  15. Google Search Central. Creating helpful, reliable, people-first content. Original information, experience, sourcing and editorial self-assessment.
  16. Google, web.dev. Build agent-friendly websites. Interface practices for browser-agent interactions.

Updated and sources checked: 19 November 2026

Last updated:

Share this post

Curious what AI could do for your business?

In a free 30-minute discovery call we look at your situation and show you where automation pays off first — no strings attached.

Book a free discovery call