Your website might help answer a customer’s question without receiving the visit. An AI response can reference your delivery guide, explain a product using your specifications or cite your advice while the customer continues researching elsewhere. That creates a new reporting question: can you see when your content is being used—and work out whether it matters to the business?

Bing’s AI Performance public preview, introduced on 10 February 2026, made some of that citation activity visible to website owners. Its coverage includes Microsoft Copilot, Bing’s AI summaries and selected partner integrations. It is not a report of every AI service. ^[1]

For Addweb’s clients, the opportunity is practical. Citation reports can help identify useful pages and questions that deserve attention. The risk is treating a new visibility number as proof of commercial success. A reference in an answer, a website session and a paid order are different observations, and a sensible report should explain each one.

What does it mean for AI to cite your website?

Some AI answers use retrieved information to support a response. Microsoft describes this retrieval process as grounding: bringing relevant external information into the answer. ^[4] A citation gives the reader a reference to supporting material. It does not, by itself, establish that the business was recommended or that anyone followed the link.

The distinction matters when interpreting a screenshot. A supplier might show your business in an AI answer and call it a ranking win. Look at what the answer actually says. Your page may support one factual sentence in a comparison that ultimately recommends another option. Accurate attribution is useful; commercial relevance still needs examination.

What changed in Bing’s reporting during 2026?

This progression makes the report more useful for investigation. A raw total can prompt a question; a relationship between a query and a page can point towards work. If the cited page is an old guide that still answers an important question, the next step might be a factual update and a clearer route to the relevant service.

The release dates also matter when comparing historical reports. Record the dates, available features and filters used in an export. If your reporting method changes, note it beside the chart. Otherwise a change in how you collect or group the evidence can be mistaken for a change in performance.

Read the metric before reacting to the number

For example, a rise in Average Cited Pages might encourage you to inspect a wider range of content. It does not tell you that more people visited those pages. Likewise, a heavily cited troubleshooting article may serve an audience with little need for your paid service. The number becomes useful when you connect it to the page’s purpose.

Start with a small set of important URLs: the core service, a valuable product category and a few supporting guides. For each, write down the customer question it should answer and the next action it should support. That gives your review a business reference point before the largest numbers attract all the attention.

Use the connection between queries and pages

Microsoft’s March update describes a view in which you can select a grounding query to inspect the associated cited pages, or select a page to inspect its related queries. The relationship can run many to many: one query can involve several pages, and one page can support several queries. ^[3]

This is useful when the wrong page appears to be doing the work. Suppose a delivery question is associated with an old promotional article rather than the current delivery policy. Inspect both pages. Check for contradictory wording, expired conditions and a missing link to the maintained policy. Correct the information because the customer needs a dependable answer, then monitor what happens.

Also look for good content that lacks a useful next step. A guide may explain a problem clearly while leaving the reader unsure whether you provide the relevant service. An appropriate service link, a clear limitation and a working enquiry route can improve the experience without turning every paragraph into a sales pitch.

Topics, Intents, Citation Share and Compare

The labels deserve care: Bing says topic and intent classifications can evolve during preview. Citation Share does not expose competitors’ domains or function as a ranking or quality score. ^[2] In practice, an intent labelled commercial should prompt investigation; it is not a record of a particular person being ready to buy.

Comparison also needs a change log. Record page edits, tracking changes, seasonal promotions and reporting changes. A rise after an edit makes the edit worth examining, but the timing alone does not prove that the edit caused the rise. Keep the comparison period and scope consistent before interpreting the movement.

How citations can increase while Citation Share falls

Bing defines Citation Share using your site’s citations divided by citations across all sites for the same grounding query. ^[2] Consider an illustrative example with two equal 30-day periods and an unchanged query and reporting scope.

In Period A, your site receives 120 citations out of 600 across all sites: 20%. In Period B, it receives 150 out of 1,000: 15%. Your own citation count rises by 25%, while your share falls by five percentage points. Both statements can be true because the denominator grew faster.

More citations can coexist with a smaller share

That is a reason to ask a better question, rather than label the month good or bad from one figure. Was the subject commercially relevant? Was the cited information correct? Did the website receive useful visits or enquiries? The share calculation supplies context for visibility; it does not answer those business questions.

Why a citation may never become a visit

A customer might get the factual answer they need, remember the business for later or continue comparing without opening the cited page. These are plausible journeys, not events that a citation report can individually verify. Some citations may therefore be useful without an immediate visit, while others may have little commercial value.

A historical study illustrates the distinction. Pew Research Center analysed browsing data from 900 US adults in March 2025. It reported that a link within an AI summary was clicked in only 1% of visits to search pages containing one. The study used later result-page snapshots to classify searches. It is a 2025 US Google observation, not a 2026 Bing benchmark or a forecast for South African businesses. ^[11]

The denominator is particularly important: that 1% describes visits to search pages with an AI summary. It is not a conversion rate for cited websites. A local service provider should establish its own evidence rather than insert that percentage into a revenue forecast.

Where useful, add an optional “How did you hear about us?” question to the enquiry process. Treat the answer as customer recollection, which may be incomplete. A person’s account, a recorded referral and a citation trend can each add context, but agreement between them still does not establish every step of the journey.

There is no single report for the whole AI search journey

Google announced its dedicated generative AI reporting in June and confirmed global availability for all websites on 31 August 2026. Its documentation states that these impressions are already included in overall performance reporting. Adding the dedicated AI total to the overall total would double-count them. ^[5]

Build a report with separate labelled panels, rather than one combined “AI visibility” total. Preserve the provider, metric, time period and filters. A citation count and an impression count may both be useful, but adding them produces a number without a consistent meaning.

A practical measurement setup for WordPress and Elementor

Start with one customer journey that matters: a completed service enquiry, an accepted booking or a paid order. Write down what counts as success and where its reliable record lives. For an enquiry, that may be the form record plus receipt by the responsible team; for a sale, it is the relevant order and payment record.

1. Inspect acquisition. GA4’s Traffic acquisition report describes sessions from new and returning users. Use session-scoped dimensions such as Session source / medium to inspect how recorded visits arrived. ^[6] Check the values actually appearing in your property before defining an “AI traffic” group. A broad domain match can accidentally include visits unrelated to AI answers.

2. Define a meaningful action. Google describes key events as events that measure actions important to a business. ^[7] For a lead-generation site, track the successful completion you intend to measure. A click on “Send” is insufficient if the form can fail validation or delivery. A WhatsApp link click is evidence of a click, not a confirmed conversation.

3. Test the published journey. Google recommends Realtime and DebugView for checking event collection. ^[8] On the actual WordPress page, submit a valid test enquiry, check that the expected event arrives and verify receipt by the business. Also try an invalid submission. It should not be recorded as a completed enquiry. Check for duplicate collection where a plugin and tag manager both add tracking.

4. Review quality outside the event count. Have the team distinguish genuine enquiries, suitable prospects and accepted work. Agree how repeat submissions and spam are handled. A higher event count can otherwise conceal lower lead quality, and one customer can generate several events.

5. Keep the rate’s denominator clear. GA4’s Session key event rate is the percentage of sessions with at least one key event. ^[6] That differs from dividing every recorded event by sessions. For a small business with few enquiries, show the underlying counts beside any percentage so a single additional lead does not look like an established trend.

Attribution is not the same as proof of cause

Google defines attribution as assigning credit to touchpoints along the path to an important action. ^[9] That is a useful reporting operation, but its output depends on what was observed and how credit was assigned. It does not automatically show whether the sale would have happened without a particular AI interaction.

Suppose a customer reads an answer, returns to your site by typing the address and phones the business. Your team may learn about the AI interaction during the call; the website record may show a different arrival source. Keep both observations, with their origins, rather than quietly rewriting one to make a perfectly connected story.

If a campaign or content project is important enough to justify closer evaluation, define the question before starting. Record the baseline, the pages being changed, the timing and other major activity. Decide what improvement would count as useful. Even when a controlled experiment is impractical, this makes the eventual conclusion more honest and actionable.

Turn the dashboard into a monthly review

A practical review process

Use a short review with a named owner. First record what changed and whether the comparison is valid. Then inspect a few relevant pages and the customer task each supports. Bring in acquisition data and the team’s enquiry or sales records. Finish with one or two decisions, an owner and a date to review the result.

For each decision, record the reason. “Correct the outdated delivery condition” is more useful than “optimise for AI”. “Make the enquiry confirmation reliable” can be verified. A vague instruction to increase an AI score leaves the developer, content editor and business owner with different ideas of success.

Keep a dated copy of the report or export used for the decision, along with its filters. If labels, coverage or your own grouping rules later change, the earlier record helps explain why two apparently similar monthly summaries do not match.

A small retailer: what useful action could look like

Consider a fictional WooCommerce retailer selling replacement parts. A compatibility guide attracts citation activity, but customers still send incomplete enquiries. The owner wants better leads, not simply more references to the guide.

Reviewing the page reveals two practical gaps: the examples do not explain which product details the team needs to check, and the enquiry link drops the reader into a generic contact page. The retailer verifies the guide against reliable product information, explains what to provide and adds an enquiry route that retains the relevant context.

The business then checks successful submissions, whether the team receives the needed details and how many requests it can actually help with. Citation trends remain part of the review. They are not the only success criterion. If enquiry quality improves while citations stay flat, the website change may still have done valuable work.

For updated pages, check whether your existing setup supports IndexNow. The protocol notifies participating search engines about added, changed or deleted URLs; it does not guarantee immediate indexing. ^[12] Confirm the current integration before adding another plugin, and do not treat a submitted URL as proof of an AI citation.

What to ask about an “AI visibility” score

Commercial monitoring tools can help organise repeated checks. Before buying one, ask what its score measures: sampled answers to a chosen prompt set, brand mentions, linked citations or something else. Those observations answer different questions. A polished percentage needs a definition and a reproducible collection method.

Ask which providers and models are tested, how prompts are selected, which countries or languages are represented, how often checks run and how missing results are handled. Request examples behind the score. If the tracked prompt list changes, ask whether the comparison remains valid. Make sure commercially important questions are included, rather than accepting an impressive total built from irrelevant topics.

Google’s guidance on third-party SEO tools says they do not have access to its internal ranking systems, and their recommendations are not Google endorsements. ^[13] Apply a similar evidential standard when evaluating AI-visibility promises: ask what is observed, what is inferred and what remains unknown. Avoid treating a supplier’s proprietary metric as an official platform measure.

Where this could lead in 2027

The documented 2026 releases show reporting becoming more specific: visibility can be examined by page, subject, query context and time. Our expectation for 2027 is that businesses will ask harder questions about the value of that visibility and the quality of the handover to their websites. This is an editorial outlook, not an announced universal reporting standard.

A useful direction would be clearer metric definitions and better connections between available reports. It would still leave gaps where people research without clicking, switch devices or contact a business offline. More detailed dashboards should help business owners express those limits more precisely, rather than encourage unsupported certainty.

Prepare by keeping a small, well-defined measurement system: important pages, reliable events, verified business outcomes and a record of changes. That gives you something stable against which to assess new reports, tools and optimisation services as they appear.

Questions business owners are asking

Can I see every time an AI tool mentions my business?

No. A citation report has a defined scope, and brand mentions are a different observation. Use the available reports for the surfaces they cover and describe any additional prompt monitoring as a sample, not a complete record of the AI web.

Does a citation mean the AI recommends us?

No. The page may support a fact, a comparison or a qualification. Inspect the answer where possible and distinguish an accurate source reference from an endorsement or a suggested purchase.

Why do I see citations but no identifiable AI visits?

The cited link may not have been followed. The visit may also fall outside your measured period or may not carry an identifiable source. Investigate the tracking and reporting scope before concluding either that the citations were worthless or that analytics must be broken.

Is Citation Share the same as market share?

No. It describes a share within a defined citation comparison. It does not measure your share of customers, revenue or an entire market. Keep the underlying query and scope attached to the number whenever you report it.

Do zero reported citations mean my site is invisible to all AI?

No. They mean the particular report has not shown citation activity for the selected scope and period. Check the property, dates, available data and coverage before making a broader claim.

Should we rewrite every page to win more citations?

Begin with evidence of a real problem: missing information, outdated facts, unclear relationships or a poor customer journey. Prioritise pages important to the business. Measure the effect of a purposeful change before extending the work across the site.

Do we need a new WordPress website or another analytics plugin?

Audit the current setup first. Useful tracking and content improvements may be possible with the existing theme, Elementor layout and analytics configuration. Add tools only for a defined gap, and check that they do not duplicate collection.

What should go in our monthly management report?

Include the relevant platform visibility measures, observable website acquisition, meaningful completed actions and verified business outcomes. Add the most important change, its likely explanations, the uncertainty and the next action. Show counts alongside percentages.

An AI citation is a useful signal when it leads to a better question. Which information is being used? Is it correct? Does the page help the right customer progress? Did the business receive a useful result? Keeping those questions separate makes the answers more valuable.