Answer
What can AI do with your website that is actually useful?
Answer the questions your enquiry form receives, find what your pages do not say, and keep the factual details consistent.
Three things: answer the questions people actually send you before they send them, find what your pages fail to say by comparing enquiries against content, and keep prices, hours and service areas consistent across every page.
The applications that get built are usually a chat widget and a content generator, and both are the weakest options available. The stronger ones start from a resource the business already has and does not use: the questions people send. Every enquiry that asks something is evidence that a page did not answer it, and the accumulated set is a precise specification for what the site is missing.
So the first useful application is analysis rather than generation. Take a period of enquiries, extract what each was asking, group them, and compare against what the site says. The output is a short ranked list of gaps, each of which is a page or a paragraph that will reduce enquiries of that type. This is more valuable than any volume of generated content, because it is grounded in demand rather than in a keyword tool.
The second is consistency. Prices, opening hours, service areas, qualifications, contact details and turnaround times accumulate across pages and drift, and the inconsistency is visible to customers and to the systems that read your site. Checking every factual detail across every page against a single source is tedious for a person and straightforward to automate, and the errors it finds are usually real.
The third is answering before the enquiry. A page that directly answers the question people keep asking removes work as well as improving the site, and it is the specific mechanism by which content reduces cost rather than only generating leads. This is why the analysis comes first: writing pages about what you assume people want has a much weaker relationship to enquiry volume than writing about what they actually asked.
A chat interface is worth a comment because it is the default suggestion. It answers a question the visitor was going to ask anyway, at the cost of a system that can be wrong on your behalf, and it does not fix the page that failed to answer. Where it earns its place is on sites with a genuinely large answer surface — many products, complex eligibility, extensive documentation — and it is a poor first investment for a business with twelve pages.
The one to be careful about is generated content at volume. Publishing many similar pages produced from the same brief is the pattern that search systems specifically penalise, and it is also the pattern that produces a site nobody wants to read. The distinction that matters is whether each page answers a distinct question that somebody actually has, which returns to the enquiry analysis and is a much smaller number of pages than a content plan usually contains.
Your enquiry inbox is a list of everything your website failed to say, and it is the only content brief you need.
Siddharth Sharma, Context Theory
Related questions
Should you write pages specifically for AI answer systems?
Write pages that answer a question directly and completely, which is what those systems extract and what readers want. The optimisation that works is structural clarity — the answer near the top, the specifics stated, the source of any figure given — and it is the same thing that makes a page useful to a person. Content written for machines and not for readers tends to satisfy neither.
Can AI tell you why your site is not converting?
It can identify what your pages do not say and compare them against what enquiries ask, which is genuinely useful. It cannot tell you why a specific visitor left, because that information does not exist in anything it can read. Treat proposed explanations as hypotheses to test rather than as diagnoses.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Ranking loss for scaled near-identical page farms | 60–90% | Category-wide |
| AI-cited sources that also rank in the Google organic top 10 | 10% | Category-wide |
Google March 2026 core update — scaled content abuse · verified
2026 generative engine citation study · fewer than · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Workflow | Every enquiry containing a question is evidence that a page failed to answer it, so the accumulated enquiry set is a demand-grounded specification for missing content that no keyword research replicates. | Extracting the questions from a period of enquiries and checking each against the site's existing pages. |
| Software | Factual details such as prices, hours, service areas and turnaround times drift across pages over time, and checking them all against a single source is tedious manually and mechanical to automate, with the discovered inconsistencies typically being real. | Comparing every stated price and opening time across the site against the current authoritative values. |
| Buying behaviour | A chat interface answers a question the visitor was going to ask anyway while leaving the unanswering page unchanged, so it earns its cost only where the answer surface is genuinely large. | Counting the distinct questions a site's content must cover before deciding whether an interface is needed to navigate them. |
| Constraint | Publishing many similar pages generated from one brief is the pattern search systems penalise, so the distinguishing test is whether each page answers a distinct question somebody actually asked. | Comparing a proposed content plan against the distinct questions found in real enquiries. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one