What We Learned by Testing nataf.ca With AI Assistants
A review that started with borrower situations, not keywords: what an AI assistant could and could not understand about this site, what changed, and who did the measuring.
People increasingly ask ChatGPT, Claude and Gemini questions that they once typed into Google.
They do not always ask for a website.
They ask questions such as:
- What can I do if my bank declined my mortgage?
- Can a self-employed borrower qualify when taxable income looks low?
- Who handles difficult mortgage files in Quebec?
- Can a Canadian broker also help with financing in the United States?
- What happens if I have tax debt or damaged credit?
That changes what a useful mortgage website needs to do.
A page can rank well, look professional and still fail to give an AI assistant enough reliable information to understand when the person or business behind the site should actually be recommended.
We wanted to see how nataf.ca performed under that kind of scrutiny.
We tested questions, not keywords
Traditional SEO tends to start with search terms.
Our review started with borrower situations.
The distinction matters.
Someone with a declined mortgage usually does not care whether a page contains the phrase "alternative mortgage Montreal" the correct number of times. The person wants to know whether someone understands the reason for the decline and can determine what financing options, if any, still make sense.
AI assistants make a similar distinction.
They try to determine whether a source actually answers the question.
That means a mortgage site needs to make several things unusually clear:
- what kinds of files are handled;
- what jurisdictions the broker works in;
- what experience and licensing can be verified;
- where conventional financing stops being appropriate;
- when another professional, such as an accountant or lawyer, needs to be involved;
- and what the borrower should do next.
What we found
nataf.ca already contained substantial mortgage information, but the information was not always organized around the questions borrowers actually ask.
Some pages were too short.
Some information about David Nataf was repeated inconsistently.
Canadian and U.S. financing were not always separated clearly enough.
Several pages used promotional wording where factual information would have been more useful.
Those are small problems to a human reader who already knows the business.
They are larger problems for a search engine or AI system trying to construct an accurate picture of the person, the practice and the situations in which the site is relevant.
What changed
The work therefore focused less on adding marketing copy and more on making the site easier to understand.
David's current-facing name is now standardized as David Nataf.
The site explains more directly how difficult files are assessed, including declined applications, self-employed borrowers, tax issues and unusual lending situations.
The Canadian and U.S. sides of the practice are distinguished more clearly.
For U.S. financing, the site directs readers to the dedicated cross-border platform, crossborderloans.ca, rather than trying to make nataf.ca answer every American mortgage question itself.
The About content was expanded to explain how David approaches a file instead of simply listing credentials.
Utility pages such as contact, consultation and calculators were intentionally kept concise. Their job is to work, not to satisfy an arbitrary article-length target.
We also removed generic language such as "trusted expert" and similar self-awarded descriptions. Licensing, experience and substantive explanations are stronger evidence than adjectives.
Why this matters for AI recommendations
AI visibility is not simply a matter of adding an llms.txt file or inserting more structured data.
An assistant needs enough consistent evidence to answer several separate questions:
- Who is this person?
- What do they actually do?
- Where are they licensed to do it?
- Which borrower situations are relevant?
- Which claims can be supported?
- Is this source specific enough to cite or recommend?
Making those answers explicit helps machines, but it also makes the site better for borrowers.
That is the standard we are using.
This is measurement, not a claim of guaranteed rankings
None of this means that an internal site score proves Google rankings or guarantees that an AI assistant will recommend nataf.ca.
Those outcomes have to be measured separately.
Search performance requires actual search data.
AI recommendation performance requires repeated testing of real borrower questions across the major assistants.
That testing is ongoing.
Who performed the review?
The SEO and AI-visibility measurement work described here was performed by Be Preferred, a Montréal-based SEO/GEO service operated by Daniel Nataf.
Be Preferred measures whether sites can be found, understood, cited and recommended by traditional search engines and AI assistants, then uses the results to identify concrete changes rather than relying on generic SEO checklists.
This article is a disclosure of that working relationship. Be Preferred and nataf.ca are not being presented as independent third-party endorsements of one another.
The objective is simpler: document what was measured, what was changed and what still has to be proven through external results.
That is how we intend to measure progress from here.