Can a freelancer add AI reception to an existing website without rebuilding it?

Check what the site already answers

Walk the home page as a stranger. Can someone tell what you provide, for whom, where examples live and how cooperation begins? An image-only portfolio without project context, scope or a contact route gives AI little reliable text. Google's English SEO Starter Guide asks for readable, original and maintained content that helps people and search systems understand a page. That work also helps an assistant answer from public facts, but cannot guarantee citation by any AI product.

The fictional illustrator has portfolio images and a contact form, but no explanation of commercial licensing, animation work or collaboration stages. Add a service page and FAQ covering accepted work, what clients should supply, deliverables, how rights are discussed, response window and contact route. A custom quote need not be a public fixed number. A previous case does not prove every new client gets the same price or rights.

Give important pages descriptive headings, readable text, image alternatives and an update owner. Google's helpful-content guidance asks whether readers can achieve their purpose. A visitor choosing an illustrator needs style, working scope and a path to discuss a project. 'We are creative' without process or deliverables is too thin for people and for grounded answers.

Divide responsibilities between site, chat and back office

A kakinang web widget can be embedded on an existing site using configured site key, allowed origin and agent settings. The owner need not move the portfolio, but an authorized manager must configure the connection; an arbitrary third-party page is not an approved origin. FAQs, supported uploads and website synchronization can become approved knowledge. Crawling requires an allowed reachable public source, and a completed candidate version before activation. Editing the site does not mean AI instantly has the new facts.

The original site still shows work, terms and the formal contact path. AI answers approved common questions and, with consent, organizes enquiries. The owner reads leads and conversations in the back office and decides whether to work together. Variable price, licensing and exceptional deliverables need human review; chat cannot sign or pay. If human handoff is disabled, do not promise it. Asynchronous means later response, while live requires staff. Email alerts need recipients and SMTP.

If the old site already has a form and analytics, choose a clear next action: retain the form, use an AI lead, or route by intent. Three equally prominent quote buttons with separate inboxes confuse visitors and staff. Trace a test enquiry from the page to its record, queue owner and response path. Choose what the freelancer can really maintain.

Test mobile access, privacy and answer boundaries

Open the home, project and contact pages on desktop and phone. Check whether chat hides navigation, images or submit buttons, and whether it can be closed with the keyboard visible. An image-heavy portfolio also needs text alternatives. The widget must not become the only place to find service information. Test public service scope, past work, unpublished price, custom rights and a human-contact request. Expect sourced facts, clear case boundaries, an honest unknown, human review and handoff according to configuration.

If the site runs on domains, subdomains and preview hosts, verify each permitted embedding origin and the production address. Keep test keys out of public articles. A strict site content-security policy may require the owner to allow legitimate script and connection sources, followed by a regression check. Explain why contact data is collected and take only what follow-up needs; sensitive case material belongs in a controlled later exchange.

Weekly, review unanswered questions, changes to source pages, lead follow-up and mobile obstruction. Repeated questions about commercial use mean the rights process needs clearer copy. Repeated availability questions need a maintained calendar or human answer, not an inference from an empty date. The existing site remains the public source; AI connects its verified content to accountable work.

Before installation, inventory the home, portfolio, service and contact pages, with an owner and update frequency for each. Correct an old price on the original page before synchronization; AI cannot infer that a three-year-old figure has expired. On multilingual sites, reconcile delivery and contact terms across languages. If someone asks to buy full copyright in a pictured work, refer to published rights terms or the creator rather than inferring rights from an image.

After embedding, record the pages involved, whether chat opens and closes, whether the mobile keyboard blocks input, and whether the original form still submits. Follow one consented test lead through the back office and human response; a second visitor declining contact should still read public answers. A page behind login, crawler blocking or temporary protection may not be available for site knowledge. Check reachability and candidate content before activation. Embedding guarantees neither indexing, search ranking, sales nor citation by an AI platform; the author still maintains the source text.

After a site redesign, check broken links. If a service URL changes, verify navigation, old search links, AI knowledge sources and the form confirmation page, not only whether chat opens on the home page. Record a test date and expected answer for each major entry point and repeat it after content updates. If an answer still cites a removed package, stop using that knowledge, resynchronize and review it before directing visitors there. A small acceptance record helps the freelancer or collaborator trace which website change affected intake.