Local Guides is a Smart Directory Pro feature (shipped in 1.5.1) that lets you publish long-form articles, anywhere from 1,500 to 15,000 words, where every business mentioned in the prose automatically links to its directory listing. The article is drafted by AI from your real directory data. Each business gets name in a typed anchor with hover card and JSON-LD schema. AI engines (ChatGPT, Perplexity, Gemini, Google AI Overviews) see each mention as a structured LocalBusiness citation rather than ambient prose, which makes your directory pages eligible to be cited as the source for “best restaurants in…” style answers.
Before you start
- An AI provider configured. Smart Directory → Settings → AI. Gemini, OpenAI, and Anthropic are all supported. The plan-then-section pipeline averages 8-12 calls per guide (roughly $0.05-$0.40 in tokens depending on length and provider).
- At least 30-50 listings in the categories and locations you want the guide to cover. The generator can’t write meaningfully about a category with fewer than ~10 entries.
- Listings have categories and locations assigned so the scope pickers work. Category scoping alone is fine if you don’t use locations.
- Optional but recommended: Pro+ license. The AI generator runs behind the same license gate as the rest of the AI features.
Generating your first guide, step by step
Step 1. Find the Guides menu
From the WordPress admin sidebar, hover Smart Directory and pick Guides. This screen behaves like any other WP post list, table of every guide on the site, quick-edit / trash / bulk actions.
Step 2. Add new and open the AI generator
Click Add New. The block editor opens with a panel in the sidebar titled “🪄 AI guide generator”. Four inputs:
- Topic prompt, what the guide is about, in your own words. The planner reshapes it into proper H2 sections. Examples: “Everything you need to know about moving to Milton Keynes”, “A complete guide to eating out in Bedford for under £15”, “How to choose a primary school in Buckingham”.
- Target word count, slider from 1,500 to 15,000. The planner respects the budget within ±10%. Below 1,500 the article reads thin; above 8,000 you’ll want to skim it before publishing.
- Scope: categories, which listing categories to draw from. Multi-select. Leave blank for site-wide.
- Scope: locations, same for location taxonomy.
Step 3. Click Generate
Hit Generate guide with AI. The button confirms, then locks while the pipeline runs. Three stages follow:
- Plan (5-15s), the planner returns a JSON outline: title, 5-15 H2 sections, per-section word budget, and which listing IDs each section should weave in.
- Sections (60-150s), one AI call per section, threading the assigned listings + their facts into the prompt. Each call returns ~800-2,200 words of HTML.
- Stitch + auto-link (under 1s), sections concatenate, the AutoLinker scans for known business names, wraps first occurrence per H2 section in a typed anchor, and records the mention set.
max_execution_time), drop the target word count and try again. Each section call is independent, partial failures bail cleanly without saving a half-finished draft.Step 4. Review the generated draft
When generation completes you’ll see the editor populated with the full draft and a green success pill in the sidebar telling you the word count and how many listings were linked. The “Guide metadata” panel below shows word count, generation timestamp, the linked-listing list (with edit links), and the section count.
Edit anything that needs editing. The AutoLinker runs again every time you save the post, so if you add a paragraph that mentions a new business, it gets wrapped automatically. Add a topic in the standard WP taxonomy sidebar (e.g. moving, food, family) so the published guide surfaces in the related-guides rail at the bottom of every other guide tagged with the same topic.
Step 5. Publish
Publish like any post. The single-guide template renders with: hero with topic chips, byline, “X min read” estimate, featured image; sticky table of contents with scroll-spy; body prose with every business mention rendered as a hover-card-equipped link; “Listings featured in this guide” mosaic; “You might be interested in” related-guides rail; Article + mentions: [LocalBusiness, …] JSON-LD in <head>.
How the auto-linker decides what to link
The auto-linker is the unglamorous-but-critical piece. Three rules govern it:
- First occurrence per H2 section. If “Wagamama” appears five times in the “Where to eat” section, only the first instance is linked. The dedup resets at every
<h2>, mimicking Wikipedia’s linking pattern. - Title match, with sensible aliases. A listing titled “John Lewis & Partners Ltd” also matches “John Lewis & Partners” and “John Lewis” (suffix stripping). “Wagamama – centre:mk” also matches just “Wagamama”.
- Place-name collision filter. If a listing happens to be titled “Milton Keynes”, the auto-linker won’t wrap every prose mention of the city to point to it. Listing titles that match an
sdp_locationterm name are excluded, along with a small stopword list of generic place phrases.
Narrowing the Features facet per category
Adjacent feature shipped in the same release: per-category facility allowlists. When a user filters by a category on a search or archive page, the “Features” facet narrows to just the facilities admins flagged as relevant. WiFi / Parking surface on Restaurants; Wheelchair access / Wedding-friendly surface on Wedding Venues.
Configure on each category term: Listings → Categories → Edit category. New field at the bottom: Relevant features, comma-separated.
Leave blank to show every facility present in the result set (default behaviour). Setting an allowlist makes the Features facet narrower and more useful at the cost of hiding long-tail facilities.
Writing a guide manually
You can write a guide entirely by hand, skip the generator, just type into the block editor. Every time you save, the AutoLinker runs over the body, wraps any business names it recognises in proper anchors, updates the mention set, refreshes the table of contents from your H2/H3 headings, and recalculates word count.
The AutoLinker is idempotent: it strips any existing <a class="sdp-guide-link"> wrappers before re-applying, so you’ll never end up with double-wrapped anchors.
The schema markup we emit
Every published guide outputs an Article JSON-LD block in <head>, with the mention set materialised as a mentions array of LocalBusiness references. Each business mention includes its canonical URL, name, image, and aggregate rating (if reviews are present). Sample output:
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "The ultimate guide to moving to Milton Keynes",
"datePublished": "2026-05-30T11:24:00+01:00",
"dateModified": "2026-05-30T13:08:00+01:00",
"author": { "@type": "Person", "name": "Editorial" },
"publisher": { "@type": "Organization", "name": "Milton Keynes Directory" },
"mainEntityOfPage": "https://mk.directory/guides/moving-to-milton-keynes/",
"image": "https://…/cover.jpg",
"mentions": [
{
"@type": "LocalBusiness",
"@id": "https://mk.directory/listing/wagamama-centre-mk/#listing",
"name": "Wagamama - centre:mk",
"url": "https://mk.directory/listing/wagamama-centre-mk/",
"image": "https://…/wagamama.jpg",
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": 4.5,
"reviewCount": 248
}
},
{ "@type": "LocalBusiness", "@id": "…#listing", "name": "Pizza Hut", "url": "…", … }
]
}
What this gets you: the guide becomes eligible to be cited by AI engines as the source for queries like “best places to eat in Milton Keynes”, and the same models see each business mention as a structured entity they can attribute, score, and link back to. Plain-prose competitors lose this lever entirely.
Topic prompts that produce good guides
The planner does its best work with topics that have an obvious structure. A few patterns that consistently produce strong drafts:
- Lifestyle umbrella topics, “Moving to {town}”, “A weekend in {town}”, “Starting a family in {town}”. Segments naturally into housing, schools, food, things to do.
- Tight verticals with personality, “Where to eat brunch in Milton Keynes”, “Best independent bookshops near Bedford”.
- How-to with local angle, “How to find a builder in Buckingham”, “Choosing a primary school in Newport Pagnell”.
Patterns that produce weaker drafts: single-business topics, off-directory topics (history, geography), or overly broad ones (a guide that wants to cover the entire UK in 8,000 words).
FAQ
Will Google flag this as AI content?
No, what Google penalises is spammy AI content built to manipulate rankings. A guide that links to dozens of real local businesses, carries structured mentions schema, and helps readers find genuinely useful services is exactly the long-tail editorial content Google’s guidelines say it wants. Review every guide before publishing.
Can I edit a generated guide?
Yes, freely. The output is plain Gutenberg blocks. The AutoLinker re-runs on every save, so new business mentions get wrapped automatically. To disable auto-linking on a specific paragraph, wrap it in a Custom HTML block, the linker skips anything inside <code>, <pre>, or existing anchors.
Can I regenerate part of a guide?
Not today. The generator runs full-document plan→sections. If you don’t like one section, deleting it from the editor and writing a replacement is the fastest path. A “regenerate this section” button is on the roadmap.
How does this differ from the SEO Pages feature?
SEO Pages are short, listicle-shaped, “Top 5 best {category} in {location}”, built for high-volume programmatic generation. Guides are long-form, editorial, individually planned: 1,500-15,000 words each, with their own outline and prose style. Both link to listings; both emit schema. Think of SEO Pages as the long-tail and Guides as the headlines.
Where do the listing facts in the prompt come from?
The generator queries your real directory data: title, primary category, primary location, average rating, review count, top six facilities, and the highest-rated approved review snippet (truncated to 220 chars). No external API or scraping, everything stays on your own database.